xref: /linux/drivers/net/ethernet/ibm/ibmvnic.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /**************************************************************************/
3 /*                                                                        */
4 /*  IBM System i and System p Virtual NIC Device Driver                   */
5 /*  Copyright (C) 2014 IBM Corp.                                          */
6 /*  Santiago Leon (santi_leon@yahoo.com)                                  */
7 /*  Thomas Falcon (tlfalcon@linux.vnet.ibm.com)                           */
8 /*  John Allen (jallen@linux.vnet.ibm.com)                                */
9 /*                                                                        */
10 /*                                                                        */
11 /* This module contains the implementation of a virtual ethernet device   */
12 /* for use with IBM i/p Series LPAR Linux. It utilizes the logical LAN    */
13 /* option of the RS/6000 Platform Architecture to interface with virtual  */
14 /* ethernet NICs that are presented to the partition by the hypervisor.   */
15 /*									   */
16 /* Messages are passed between the VNIC driver and the VNIC server using  */
17 /* Command/Response Queues (CRQs) and sub CRQs (sCRQs). CRQs are used to  */
18 /* issue and receive commands that initiate communication with the server */
19 /* on driver initialization. Sub CRQs (sCRQs) are similar to CRQs, but    */
20 /* are used by the driver to notify the server that a packet is           */
21 /* ready for transmission or that a buffer has been added to receive a    */
22 /* packet. Subsequently, sCRQs are used by the server to notify the       */
23 /* driver that a packet transmission has been completed or that a packet  */
24 /* has been received and placed in a waiting buffer.                      */
25 /*                                                                        */
26 /* In lieu of a more conventional "on-the-fly" DMA mapping strategy in    */
27 /* which skbs are DMA mapped and immediately unmapped when the transmit   */
28 /* or receive has been completed, the VNIC driver is required to use      */
29 /* "long term mapping". This entails that large, continuous DMA mapped    */
30 /* buffers are allocated on driver initialization and these buffers are   */
31 /* then continuously reused to pass skbs to and from the VNIC server.     */
32 /*                                                                        */
33 /**************************************************************************/
34 
35 #include <linux/module.h>
36 #include <linux/moduleparam.h>
37 #include <linux/types.h>
38 #include <linux/errno.h>
39 #include <linux/completion.h>
40 #include <linux/ioport.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kernel.h>
43 #include <linux/netdevice.h>
44 #include <linux/etherdevice.h>
45 #include <linux/skbuff.h>
46 #include <linux/init.h>
47 #include <linux/delay.h>
48 #include <linux/mm.h>
49 #include <linux/ethtool.h>
50 #include <linux/proc_fs.h>
51 #include <linux/if_arp.h>
52 #include <linux/in.h>
53 #include <linux/ip.h>
54 #include <linux/ipv6.h>
55 #include <linux/irq.h>
56 #include <linux/irqdomain.h>
57 #include <linux/kthread.h>
58 #include <linux/seq_file.h>
59 #include <linux/interrupt.h>
60 #include <net/net_namespace.h>
61 #include <asm/hvcall.h>
62 #include <linux/atomic.h>
63 #include <asm/vio.h>
64 #include <asm/xive.h>
65 #include <asm/iommu.h>
66 #include <linux/uaccess.h>
67 #include <asm/firmware.h>
68 #include <linux/workqueue.h>
69 #include <linux/if_vlan.h>
70 #include <linux/utsname.h>
71 #include <linux/cpu.h>
72 
73 #include "ibmvnic.h"
74 
75 static const char ibmvnic_driver_name[] = "ibmvnic";
76 static const char ibmvnic_driver_string[] = "IBM System i/p Virtual NIC Driver";
77 
78 MODULE_AUTHOR("Santiago Leon");
79 MODULE_DESCRIPTION("IBM System i/p Virtual NIC Driver");
80 MODULE_LICENSE("GPL");
81 MODULE_VERSION(IBMVNIC_DRIVER_VERSION);
82 
83 static int ibmvnic_version = IBMVNIC_INITIAL_VERSION;
84 static void release_sub_crqs(struct ibmvnic_adapter *, bool);
85 static int ibmvnic_reset_crq(struct ibmvnic_adapter *);
86 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *);
87 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *);
88 static int ibmvnic_send_crq(struct ibmvnic_adapter *, union ibmvnic_crq *);
89 static int send_subcrq_indirect(struct ibmvnic_adapter *, u64, u64, u64);
90 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance);
91 static int enable_scrq_irq(struct ibmvnic_adapter *,
92 			   struct ibmvnic_sub_crq_queue *);
93 static int disable_scrq_irq(struct ibmvnic_adapter *,
94 			    struct ibmvnic_sub_crq_queue *);
95 static int pending_scrq(struct ibmvnic_adapter *,
96 			struct ibmvnic_sub_crq_queue *);
97 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *,
98 					struct ibmvnic_sub_crq_queue *);
99 static int ibmvnic_poll(struct napi_struct *napi, int data);
100 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter);
101 static inline void reinit_init_done(struct ibmvnic_adapter *adapter);
102 static void send_query_map(struct ibmvnic_adapter *adapter);
103 static int send_request_map(struct ibmvnic_adapter *, dma_addr_t, u32, u8);
104 static int send_request_unmap(struct ibmvnic_adapter *, u8);
105 static int send_login(struct ibmvnic_adapter *adapter);
106 static void send_query_cap(struct ibmvnic_adapter *adapter);
107 static int init_sub_crqs(struct ibmvnic_adapter *);
108 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter);
109 static int ibmvnic_reset_init(struct ibmvnic_adapter *, bool reset);
110 static void release_crq_queue(struct ibmvnic_adapter *);
111 static int __ibmvnic_set_mac(struct net_device *, u8 *);
112 static int init_crq_queue(struct ibmvnic_adapter *adapter);
113 static int send_query_phys_parms(struct ibmvnic_adapter *adapter);
114 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
115 					 struct ibmvnic_sub_crq_queue *tx_scrq);
116 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
117 				struct ibmvnic_long_term_buff *ltb);
118 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter);
119 static void flush_reset_queue(struct ibmvnic_adapter *adapter);
120 static void print_subcrq_error(struct device *dev, int rc, const char *func);
121 
122 struct ibmvnic_stat {
123 	char name[ETH_GSTRING_LEN];
124 	int offset;
125 };
126 
127 #define IBMVNIC_STAT_OFF(stat) (offsetof(struct ibmvnic_adapter, stats) + \
128 			     offsetof(struct ibmvnic_statistics, stat))
129 #define IBMVNIC_GET_STAT(a, off) (*((u64 *)(((unsigned long)(a)) + (off))))
130 
131 static const struct ibmvnic_stat ibmvnic_stats[] = {
132 	{"rx_packets", IBMVNIC_STAT_OFF(rx_packets)},
133 	{"rx_bytes", IBMVNIC_STAT_OFF(rx_bytes)},
134 	{"tx_packets", IBMVNIC_STAT_OFF(tx_packets)},
135 	{"tx_bytes", IBMVNIC_STAT_OFF(tx_bytes)},
136 	{"ucast_tx_packets", IBMVNIC_STAT_OFF(ucast_tx_packets)},
137 	{"ucast_rx_packets", IBMVNIC_STAT_OFF(ucast_rx_packets)},
138 	{"mcast_tx_packets", IBMVNIC_STAT_OFF(mcast_tx_packets)},
139 	{"mcast_rx_packets", IBMVNIC_STAT_OFF(mcast_rx_packets)},
140 	{"bcast_tx_packets", IBMVNIC_STAT_OFF(bcast_tx_packets)},
141 	{"bcast_rx_packets", IBMVNIC_STAT_OFF(bcast_rx_packets)},
142 	{"align_errors", IBMVNIC_STAT_OFF(align_errors)},
143 	{"fcs_errors", IBMVNIC_STAT_OFF(fcs_errors)},
144 	{"single_collision_frames", IBMVNIC_STAT_OFF(single_collision_frames)},
145 	{"multi_collision_frames", IBMVNIC_STAT_OFF(multi_collision_frames)},
146 	{"sqe_test_errors", IBMVNIC_STAT_OFF(sqe_test_errors)},
147 	{"deferred_tx", IBMVNIC_STAT_OFF(deferred_tx)},
148 	{"late_collisions", IBMVNIC_STAT_OFF(late_collisions)},
149 	{"excess_collisions", IBMVNIC_STAT_OFF(excess_collisions)},
150 	{"internal_mac_tx_errors", IBMVNIC_STAT_OFF(internal_mac_tx_errors)},
151 	{"carrier_sense", IBMVNIC_STAT_OFF(carrier_sense)},
152 	{"too_long_frames", IBMVNIC_STAT_OFF(too_long_frames)},
153 	{"internal_mac_rx_errors", IBMVNIC_STAT_OFF(internal_mac_rx_errors)},
154 };
155 
156 static int send_crq_init_complete(struct ibmvnic_adapter *adapter)
157 {
158 	union ibmvnic_crq crq;
159 
160 	memset(&crq, 0, sizeof(crq));
161 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
162 	crq.generic.cmd = IBMVNIC_CRQ_INIT_COMPLETE;
163 
164 	return ibmvnic_send_crq(adapter, &crq);
165 }
166 
167 static int send_version_xchg(struct ibmvnic_adapter *adapter)
168 {
169 	union ibmvnic_crq crq;
170 
171 	memset(&crq, 0, sizeof(crq));
172 	crq.version_exchange.first = IBMVNIC_CRQ_CMD;
173 	crq.version_exchange.cmd = VERSION_EXCHANGE;
174 	crq.version_exchange.version = cpu_to_be16(ibmvnic_version);
175 
176 	return ibmvnic_send_crq(adapter, &crq);
177 }
178 
179 static void ibmvnic_clean_queue_affinity(struct ibmvnic_adapter *adapter,
180 					 struct ibmvnic_sub_crq_queue *queue)
181 {
182 	if (!(queue && queue->irq))
183 		return;
184 
185 	cpumask_clear(queue->affinity_mask);
186 
187 	if (irq_set_affinity_and_hint(queue->irq, NULL))
188 		netdev_warn(adapter->netdev,
189 			    "%s: Clear affinity failed, queue addr = %p, IRQ = %d\n",
190 			    __func__, queue, queue->irq);
191 }
192 
193 static void ibmvnic_clean_affinity(struct ibmvnic_adapter *adapter)
194 {
195 	struct ibmvnic_sub_crq_queue **rxqs;
196 	struct ibmvnic_sub_crq_queue **txqs;
197 	int num_rxqs, num_txqs;
198 	int i;
199 
200 	rxqs = adapter->rx_scrq;
201 	txqs = adapter->tx_scrq;
202 	num_txqs = adapter->num_active_tx_scrqs;
203 	num_rxqs = adapter->num_active_rx_scrqs;
204 
205 	netdev_dbg(adapter->netdev, "%s: Cleaning irq affinity hints", __func__);
206 	if (txqs) {
207 		for (i = 0; i < num_txqs; i++)
208 			ibmvnic_clean_queue_affinity(adapter, txqs[i]);
209 	}
210 	if (rxqs) {
211 		for (i = 0; i < num_rxqs; i++)
212 			ibmvnic_clean_queue_affinity(adapter, rxqs[i]);
213 	}
214 }
215 
216 static int ibmvnic_set_queue_affinity(struct ibmvnic_sub_crq_queue *queue,
217 				      unsigned int *cpu, int *stragglers,
218 				      int stride)
219 {
220 	cpumask_var_t mask;
221 	int i;
222 	int rc = 0;
223 
224 	if (!(queue && queue->irq))
225 		return rc;
226 
227 	/* cpumask_var_t is either a pointer or array, allocation works here */
228 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
229 		return -ENOMEM;
230 
231 	/* while we have extra cpu give one extra to this irq */
232 	if (*stragglers) {
233 		stride++;
234 		(*stragglers)--;
235 	}
236 	/* atomic write is safer than writing bit by bit directly */
237 	for (i = 0; i < stride; i++) {
238 		cpumask_set_cpu(*cpu, mask);
239 		*cpu = cpumask_next_wrap(*cpu, cpu_online_mask,
240 					 nr_cpu_ids, false);
241 	}
242 	/* set queue affinity mask */
243 	cpumask_copy(queue->affinity_mask, mask);
244 	rc = irq_set_affinity_and_hint(queue->irq, queue->affinity_mask);
245 	free_cpumask_var(mask);
246 
247 	return rc;
248 }
249 
250 /* assumes cpu read lock is held */
251 static void ibmvnic_set_affinity(struct ibmvnic_adapter *adapter)
252 {
253 	struct ibmvnic_sub_crq_queue **rxqs = adapter->rx_scrq;
254 	struct ibmvnic_sub_crq_queue **txqs = adapter->tx_scrq;
255 	struct ibmvnic_sub_crq_queue *queue;
256 	int num_rxqs = adapter->num_active_rx_scrqs, i_rxqs = 0;
257 	int num_txqs = adapter->num_active_tx_scrqs, i_txqs = 0;
258 	int total_queues, stride, stragglers, i;
259 	unsigned int num_cpu, cpu;
260 	bool is_rx_queue;
261 	int rc = 0;
262 
263 	netdev_dbg(adapter->netdev, "%s: Setting irq affinity hints", __func__);
264 	if (!(adapter->rx_scrq && adapter->tx_scrq)) {
265 		netdev_warn(adapter->netdev,
266 			    "%s: Set affinity failed, queues not allocated\n",
267 			    __func__);
268 		return;
269 	}
270 
271 	total_queues = num_rxqs + num_txqs;
272 	num_cpu = num_online_cpus();
273 	/* number of cpu's assigned per irq */
274 	stride = max_t(int, num_cpu / total_queues, 1);
275 	/* number of leftover cpu's */
276 	stragglers = num_cpu >= total_queues ? num_cpu % total_queues : 0;
277 	/* next available cpu to assign irq to */
278 	cpu = cpumask_next(-1, cpu_online_mask);
279 
280 	for (i = 0; i < total_queues; i++) {
281 		is_rx_queue = false;
282 		/* balance core load by alternating rx and tx assignments
283 		 * ex: TX0 -> RX0 -> TX1 -> RX1 etc.
284 		 */
285 		if ((i % 2 == 1 && i_rxqs < num_rxqs) || i_txqs == num_txqs) {
286 			queue = rxqs[i_rxqs++];
287 			is_rx_queue = true;
288 		} else {
289 			queue = txqs[i_txqs++];
290 		}
291 
292 		rc = ibmvnic_set_queue_affinity(queue, &cpu, &stragglers,
293 						stride);
294 		if (rc)
295 			goto out;
296 
297 		if (!queue || is_rx_queue)
298 			continue;
299 
300 		rc = __netif_set_xps_queue(adapter->netdev,
301 					   cpumask_bits(queue->affinity_mask),
302 					   i_txqs - 1, XPS_CPUS);
303 		if (rc)
304 			netdev_warn(adapter->netdev, "%s: Set XPS on queue %d failed, rc = %d.\n",
305 				    __func__, i_txqs - 1, rc);
306 	}
307 
308 out:
309 	if (rc) {
310 		netdev_warn(adapter->netdev,
311 			    "%s: Set affinity failed, queue addr = %p, IRQ = %d, rc = %d.\n",
312 			    __func__, queue, queue->irq, rc);
313 		ibmvnic_clean_affinity(adapter);
314 	}
315 }
316 
317 static int ibmvnic_cpu_online(unsigned int cpu, struct hlist_node *node)
318 {
319 	struct ibmvnic_adapter *adapter;
320 
321 	adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node);
322 	ibmvnic_set_affinity(adapter);
323 	return 0;
324 }
325 
326 static int ibmvnic_cpu_dead(unsigned int cpu, struct hlist_node *node)
327 {
328 	struct ibmvnic_adapter *adapter;
329 
330 	adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node_dead);
331 	ibmvnic_set_affinity(adapter);
332 	return 0;
333 }
334 
335 static int ibmvnic_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
336 {
337 	struct ibmvnic_adapter *adapter;
338 
339 	adapter = hlist_entry_safe(node, struct ibmvnic_adapter, node);
340 	ibmvnic_clean_affinity(adapter);
341 	return 0;
342 }
343 
344 static enum cpuhp_state ibmvnic_online;
345 
346 static int ibmvnic_cpu_notif_add(struct ibmvnic_adapter *adapter)
347 {
348 	int ret;
349 
350 	ret = cpuhp_state_add_instance_nocalls(ibmvnic_online, &adapter->node);
351 	if (ret)
352 		return ret;
353 	ret = cpuhp_state_add_instance_nocalls(CPUHP_IBMVNIC_DEAD,
354 					       &adapter->node_dead);
355 	if (!ret)
356 		return ret;
357 	cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node);
358 	return ret;
359 }
360 
361 static void ibmvnic_cpu_notif_remove(struct ibmvnic_adapter *adapter)
362 {
363 	cpuhp_state_remove_instance_nocalls(ibmvnic_online, &adapter->node);
364 	cpuhp_state_remove_instance_nocalls(CPUHP_IBMVNIC_DEAD,
365 					    &adapter->node_dead);
366 }
367 
368 static long h_reg_sub_crq(unsigned long unit_address, unsigned long token,
369 			  unsigned long length, unsigned long *number,
370 			  unsigned long *irq)
371 {
372 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
373 	long rc;
374 
375 	rc = plpar_hcall(H_REG_SUB_CRQ, retbuf, unit_address, token, length);
376 	*number = retbuf[0];
377 	*irq = retbuf[1];
378 
379 	return rc;
380 }
381 
382 /**
383  * ibmvnic_wait_for_completion - Check device state and wait for completion
384  * @adapter: private device data
385  * @comp_done: completion structure to wait for
386  * @timeout: time to wait in milliseconds
387  *
388  * Wait for a completion signal or until the timeout limit is reached
389  * while checking that the device is still active.
390  */
391 static int ibmvnic_wait_for_completion(struct ibmvnic_adapter *adapter,
392 				       struct completion *comp_done,
393 				       unsigned long timeout)
394 {
395 	struct net_device *netdev;
396 	unsigned long div_timeout;
397 	u8 retry;
398 
399 	netdev = adapter->netdev;
400 	retry = 5;
401 	div_timeout = msecs_to_jiffies(timeout / retry);
402 	while (true) {
403 		if (!adapter->crq.active) {
404 			netdev_err(netdev, "Device down!\n");
405 			return -ENODEV;
406 		}
407 		if (!retry--)
408 			break;
409 		if (wait_for_completion_timeout(comp_done, div_timeout))
410 			return 0;
411 	}
412 	netdev_err(netdev, "Operation timed out.\n");
413 	return -ETIMEDOUT;
414 }
415 
416 /**
417  * reuse_ltb() - Check if a long term buffer can be reused
418  * @ltb:  The long term buffer to be checked
419  * @size: The size of the long term buffer.
420  *
421  * An LTB can be reused unless its size has changed.
422  *
423  * Return: Return true if the LTB can be reused, false otherwise.
424  */
425 static bool reuse_ltb(struct ibmvnic_long_term_buff *ltb, int size)
426 {
427 	return (ltb->buff && ltb->size == size);
428 }
429 
430 /**
431  * alloc_long_term_buff() - Allocate a long term buffer (LTB)
432  *
433  * @adapter: ibmvnic adapter associated to the LTB
434  * @ltb:     container object for the LTB
435  * @size:    size of the LTB
436  *
437  * Allocate an LTB of the specified size and notify VIOS.
438  *
439  * If the given @ltb already has the correct size, reuse it. Otherwise if
440  * its non-NULL, free it. Then allocate a new one of the correct size.
441  * Notify the VIOS either way since we may now be working with a new VIOS.
442  *
443  * Allocating larger chunks of memory during resets, specially LPM or under
444  * low memory situations can cause resets to fail/timeout and for LPAR to
445  * lose connectivity. So hold onto the LTB even if we fail to communicate
446  * with the VIOS and reuse it on next open. Free LTB when adapter is closed.
447  *
448  * Return: 0 if we were able to allocate the LTB and notify the VIOS and
449  *	   a negative value otherwise.
450  */
451 static int alloc_long_term_buff(struct ibmvnic_adapter *adapter,
452 				struct ibmvnic_long_term_buff *ltb, int size)
453 {
454 	struct device *dev = &adapter->vdev->dev;
455 	u64 prev = 0;
456 	int rc;
457 
458 	if (!reuse_ltb(ltb, size)) {
459 		dev_dbg(dev,
460 			"LTB size changed from 0x%llx to 0x%x, reallocating\n",
461 			 ltb->size, size);
462 		prev = ltb->size;
463 		free_long_term_buff(adapter, ltb);
464 	}
465 
466 	if (ltb->buff) {
467 		dev_dbg(dev, "Reusing LTB [map %d, size 0x%llx]\n",
468 			ltb->map_id, ltb->size);
469 	} else {
470 		ltb->buff = dma_alloc_coherent(dev, size, &ltb->addr,
471 					       GFP_KERNEL);
472 		if (!ltb->buff) {
473 			dev_err(dev, "Couldn't alloc long term buffer\n");
474 			return -ENOMEM;
475 		}
476 		ltb->size = size;
477 
478 		ltb->map_id = find_first_zero_bit(adapter->map_ids,
479 						  MAX_MAP_ID);
480 		bitmap_set(adapter->map_ids, ltb->map_id, 1);
481 
482 		dev_dbg(dev,
483 			"Allocated new LTB [map %d, size 0x%llx was 0x%llx]\n",
484 			 ltb->map_id, ltb->size, prev);
485 	}
486 
487 	/* Ensure ltb is zeroed - specially when reusing it. */
488 	memset(ltb->buff, 0, ltb->size);
489 
490 	mutex_lock(&adapter->fw_lock);
491 	adapter->fw_done_rc = 0;
492 	reinit_completion(&adapter->fw_done);
493 
494 	rc = send_request_map(adapter, ltb->addr, ltb->size, ltb->map_id);
495 	if (rc) {
496 		dev_err(dev, "send_request_map failed, rc = %d\n", rc);
497 		goto out;
498 	}
499 
500 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
501 	if (rc) {
502 		dev_err(dev, "LTB map request aborted or timed out, rc = %d\n",
503 			rc);
504 		goto out;
505 	}
506 
507 	if (adapter->fw_done_rc) {
508 		dev_err(dev, "Couldn't map LTB, rc = %d\n",
509 			adapter->fw_done_rc);
510 		rc = -EIO;
511 		goto out;
512 	}
513 	rc = 0;
514 out:
515 	/* don't free LTB on communication error - see function header */
516 	mutex_unlock(&adapter->fw_lock);
517 	return rc;
518 }
519 
520 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
521 				struct ibmvnic_long_term_buff *ltb)
522 {
523 	struct device *dev = &adapter->vdev->dev;
524 
525 	if (!ltb->buff)
526 		return;
527 
528 	/* VIOS automatically unmaps the long term buffer at remote
529 	 * end for the following resets:
530 	 * FAILOVER, MOBILITY, TIMEOUT.
531 	 */
532 	if (adapter->reset_reason != VNIC_RESET_FAILOVER &&
533 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
534 	    adapter->reset_reason != VNIC_RESET_TIMEOUT)
535 		send_request_unmap(adapter, ltb->map_id);
536 
537 	dma_free_coherent(dev, ltb->size, ltb->buff, ltb->addr);
538 
539 	ltb->buff = NULL;
540 	/* mark this map_id free */
541 	bitmap_clear(adapter->map_ids, ltb->map_id, 1);
542 	ltb->map_id = 0;
543 }
544 
545 /**
546  * free_ltb_set - free the given set of long term buffers (LTBS)
547  * @adapter: The ibmvnic adapter containing this ltb set
548  * @ltb_set: The ltb_set to be freed
549  *
550  * Free the set of LTBs in the given set.
551  */
552 
553 static void free_ltb_set(struct ibmvnic_adapter *adapter,
554 			 struct ibmvnic_ltb_set *ltb_set)
555 {
556 	int i;
557 
558 	for (i = 0; i < ltb_set->num_ltbs; i++)
559 		free_long_term_buff(adapter, &ltb_set->ltbs[i]);
560 
561 	kfree(ltb_set->ltbs);
562 	ltb_set->ltbs = NULL;
563 	ltb_set->num_ltbs = 0;
564 }
565 
566 /**
567  * alloc_ltb_set() - Allocate a set of long term buffers (LTBs)
568  *
569  * @adapter: ibmvnic adapter associated to the LTB
570  * @ltb_set: container object for the set of LTBs
571  * @num_buffs: Number of buffers in the LTB
572  * @buff_size: Size of each buffer in the LTB
573  *
574  * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size
575  * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the
576  * new set of LTBs have fewer LTBs than the old set, free the excess LTBs.
577  * If new set needs more than in old set, allocate the remaining ones.
578  * Try and reuse as many LTBs as possible and avoid reallocation.
579  *
580  * Any changes to this allocation strategy must be reflected in
581  * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb().
582  */
583 static int alloc_ltb_set(struct ibmvnic_adapter *adapter,
584 			 struct ibmvnic_ltb_set *ltb_set, int num_buffs,
585 			 int buff_size)
586 {
587 	struct device *dev = &adapter->vdev->dev;
588 	struct ibmvnic_ltb_set old_set;
589 	struct ibmvnic_ltb_set new_set;
590 	int rem_size;
591 	int tot_size;		/* size of all ltbs */
592 	int ltb_size;		/* size of one ltb */
593 	int nltbs;
594 	int rc;
595 	int n;
596 	int i;
597 
598 	dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs,
599 		buff_size);
600 
601 	ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size);
602 	tot_size = num_buffs * buff_size;
603 
604 	if (ltb_size > tot_size)
605 		ltb_size = tot_size;
606 
607 	nltbs = tot_size / ltb_size;
608 	if (tot_size % ltb_size)
609 		nltbs++;
610 
611 	old_set = *ltb_set;
612 
613 	if (old_set.num_ltbs == nltbs) {
614 		new_set = old_set;
615 	} else {
616 		int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff);
617 
618 		new_set.ltbs = kzalloc(tmp, GFP_KERNEL);
619 		if (!new_set.ltbs)
620 			return -ENOMEM;
621 
622 		new_set.num_ltbs = nltbs;
623 
624 		/* Free any excess ltbs in old set */
625 		for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++)
626 			free_long_term_buff(adapter, &old_set.ltbs[i]);
627 
628 		/* Copy remaining ltbs to new set. All LTBs except the
629 		 * last one are of the same size. alloc_long_term_buff()
630 		 * will realloc if the size changes.
631 		 */
632 		n = min(old_set.num_ltbs, new_set.num_ltbs);
633 		for (i = 0; i < n; i++)
634 			new_set.ltbs[i] = old_set.ltbs[i];
635 
636 		/* Any additional ltbs in new set will have NULL ltbs for
637 		 * now and will be allocated in alloc_long_term_buff().
638 		 */
639 
640 		/* We no longer need the old_set so free it. Note that we
641 		 * may have reused some ltbs from old set and freed excess
642 		 * ltbs above. So we only need to free the container now
643 		 * not the LTBs themselves. (i.e. dont free_ltb_set()!)
644 		 */
645 		kfree(old_set.ltbs);
646 		old_set.ltbs = NULL;
647 		old_set.num_ltbs = 0;
648 
649 		/* Install the new set. If allocations fail below, we will
650 		 * retry later and know what size LTBs we need.
651 		 */
652 		*ltb_set = new_set;
653 	}
654 
655 	i = 0;
656 	rem_size = tot_size;
657 	while (rem_size) {
658 		if (ltb_size > rem_size)
659 			ltb_size = rem_size;
660 
661 		rem_size -= ltb_size;
662 
663 		rc = alloc_long_term_buff(adapter, &new_set.ltbs[i], ltb_size);
664 		if (rc)
665 			goto out;
666 		i++;
667 	}
668 
669 	WARN_ON(i != new_set.num_ltbs);
670 
671 	return 0;
672 out:
673 	/* We may have allocated one/more LTBs before failing and we
674 	 * want to try and reuse on next reset. So don't free ltb set.
675 	 */
676 	return rc;
677 }
678 
679 /**
680  * map_rxpool_buf_to_ltb - Map given rxpool buffer to offset in an LTB.
681  * @rxpool: The receive buffer pool containing buffer
682  * @bufidx: Index of buffer in rxpool
683  * @ltbp: (Output) pointer to the long term buffer containing the buffer
684  * @offset: (Output) offset of buffer in the LTB from @ltbp
685  *
686  * Map the given buffer identified by [rxpool, bufidx] to an LTB in the
687  * pool and its corresponding offset. Assume for now that each LTB is of
688  * different size but could possibly be optimized based on the allocation
689  * strategy in alloc_ltb_set().
690  */
691 static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool,
692 				  unsigned int bufidx,
693 				  struct ibmvnic_long_term_buff **ltbp,
694 				  unsigned int *offset)
695 {
696 	struct ibmvnic_long_term_buff *ltb;
697 	int nbufs;	/* # of buffers in one ltb */
698 	int i;
699 
700 	WARN_ON(bufidx >= rxpool->size);
701 
702 	for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) {
703 		ltb = &rxpool->ltb_set.ltbs[i];
704 		nbufs = ltb->size / rxpool->buff_size;
705 		if (bufidx < nbufs)
706 			break;
707 		bufidx -= nbufs;
708 	}
709 
710 	*ltbp = ltb;
711 	*offset = bufidx * rxpool->buff_size;
712 }
713 
714 /**
715  * map_txpool_buf_to_ltb - Map given txpool buffer to offset in an LTB.
716  * @txpool: The transmit buffer pool containing buffer
717  * @bufidx: Index of buffer in txpool
718  * @ltbp: (Output) pointer to the long term buffer (LTB) containing the buffer
719  * @offset: (Output) offset of buffer in the LTB from @ltbp
720  *
721  * Map the given buffer identified by [txpool, bufidx] to an LTB in the
722  * pool and its corresponding offset.
723  */
724 static void map_txpool_buf_to_ltb(struct ibmvnic_tx_pool *txpool,
725 				  unsigned int bufidx,
726 				  struct ibmvnic_long_term_buff **ltbp,
727 				  unsigned int *offset)
728 {
729 	struct ibmvnic_long_term_buff *ltb;
730 	int nbufs;	/* # of buffers in one ltb */
731 	int i;
732 
733 	WARN_ON_ONCE(bufidx >= txpool->num_buffers);
734 
735 	for (i = 0; i < txpool->ltb_set.num_ltbs; i++) {
736 		ltb = &txpool->ltb_set.ltbs[i];
737 		nbufs = ltb->size / txpool->buf_size;
738 		if (bufidx < nbufs)
739 			break;
740 		bufidx -= nbufs;
741 	}
742 
743 	*ltbp = ltb;
744 	*offset = bufidx * txpool->buf_size;
745 }
746 
747 static void deactivate_rx_pools(struct ibmvnic_adapter *adapter)
748 {
749 	int i;
750 
751 	for (i = 0; i < adapter->num_active_rx_pools; i++)
752 		adapter->rx_pool[i].active = 0;
753 }
754 
755 static void replenish_rx_pool(struct ibmvnic_adapter *adapter,
756 			      struct ibmvnic_rx_pool *pool)
757 {
758 	int count = pool->size - atomic_read(&pool->available);
759 	u64 handle = adapter->rx_scrq[pool->index]->handle;
760 	struct device *dev = &adapter->vdev->dev;
761 	struct ibmvnic_ind_xmit_queue *ind_bufp;
762 	struct ibmvnic_sub_crq_queue *rx_scrq;
763 	struct ibmvnic_long_term_buff *ltb;
764 	union sub_crq *sub_crq;
765 	int buffers_added = 0;
766 	unsigned long lpar_rc;
767 	struct sk_buff *skb;
768 	unsigned int offset;
769 	dma_addr_t dma_addr;
770 	unsigned char *dst;
771 	int shift = 0;
772 	int bufidx;
773 	int i;
774 
775 	if (!pool->active)
776 		return;
777 
778 	rx_scrq = adapter->rx_scrq[pool->index];
779 	ind_bufp = &rx_scrq->ind_buf;
780 
781 	/* netdev_skb_alloc() could have failed after we saved a few skbs
782 	 * in the indir_buf and we would not have sent them to VIOS yet.
783 	 * To account for them, start the loop at ind_bufp->index rather
784 	 * than 0. If we pushed all the skbs to VIOS, ind_bufp->index will
785 	 * be 0.
786 	 */
787 	for (i = ind_bufp->index; i < count; ++i) {
788 		bufidx = pool->free_map[pool->next_free];
789 
790 		/* We maybe reusing the skb from earlier resets. Allocate
791 		 * only if necessary. But since the LTB may have changed
792 		 * during reset (see init_rx_pools()), update LTB below
793 		 * even if reusing skb.
794 		 */
795 		skb = pool->rx_buff[bufidx].skb;
796 		if (!skb) {
797 			skb = netdev_alloc_skb(adapter->netdev,
798 					       pool->buff_size);
799 			if (!skb) {
800 				dev_err(dev, "Couldn't replenish rx buff\n");
801 				adapter->replenish_no_mem++;
802 				break;
803 			}
804 		}
805 
806 		pool->free_map[pool->next_free] = IBMVNIC_INVALID_MAP;
807 		pool->next_free = (pool->next_free + 1) % pool->size;
808 
809 		/* Copy the skb to the long term mapped DMA buffer */
810 		map_rxpool_buf_to_ltb(pool, bufidx, &ltb, &offset);
811 		dst = ltb->buff + offset;
812 		memset(dst, 0, pool->buff_size);
813 		dma_addr = ltb->addr + offset;
814 
815 		/* add the skb to an rx_buff in the pool */
816 		pool->rx_buff[bufidx].data = dst;
817 		pool->rx_buff[bufidx].dma = dma_addr;
818 		pool->rx_buff[bufidx].skb = skb;
819 		pool->rx_buff[bufidx].pool_index = pool->index;
820 		pool->rx_buff[bufidx].size = pool->buff_size;
821 
822 		/* queue the rx_buff for the next send_subcrq_indirect */
823 		sub_crq = &ind_bufp->indir_arr[ind_bufp->index++];
824 		memset(sub_crq, 0, sizeof(*sub_crq));
825 		sub_crq->rx_add.first = IBMVNIC_CRQ_CMD;
826 		sub_crq->rx_add.correlator =
827 		    cpu_to_be64((u64)&pool->rx_buff[bufidx]);
828 		sub_crq->rx_add.ioba = cpu_to_be32(dma_addr);
829 		sub_crq->rx_add.map_id = ltb->map_id;
830 
831 		/* The length field of the sCRQ is defined to be 24 bits so the
832 		 * buffer size needs to be left shifted by a byte before it is
833 		 * converted to big endian to prevent the last byte from being
834 		 * truncated.
835 		 */
836 #ifdef __LITTLE_ENDIAN__
837 		shift = 8;
838 #endif
839 		sub_crq->rx_add.len = cpu_to_be32(pool->buff_size << shift);
840 
841 		/* if send_subcrq_indirect queue is full, flush to VIOS */
842 		if (ind_bufp->index == IBMVNIC_MAX_IND_DESCS ||
843 		    i == count - 1) {
844 			lpar_rc =
845 				send_subcrq_indirect(adapter, handle,
846 						     (u64)ind_bufp->indir_dma,
847 						     (u64)ind_bufp->index);
848 			if (lpar_rc != H_SUCCESS)
849 				goto failure;
850 			buffers_added += ind_bufp->index;
851 			adapter->replenish_add_buff_success += ind_bufp->index;
852 			ind_bufp->index = 0;
853 		}
854 	}
855 	atomic_add(buffers_added, &pool->available);
856 	return;
857 
858 failure:
859 	if (lpar_rc != H_PARAMETER && lpar_rc != H_CLOSED)
860 		dev_err_ratelimited(dev, "rx: replenish packet buffer failed\n");
861 	for (i = ind_bufp->index - 1; i >= 0; --i) {
862 		struct ibmvnic_rx_buff *rx_buff;
863 
864 		pool->next_free = pool->next_free == 0 ?
865 				  pool->size - 1 : pool->next_free - 1;
866 		sub_crq = &ind_bufp->indir_arr[i];
867 		rx_buff = (struct ibmvnic_rx_buff *)
868 				be64_to_cpu(sub_crq->rx_add.correlator);
869 		bufidx = (int)(rx_buff - pool->rx_buff);
870 		pool->free_map[pool->next_free] = bufidx;
871 		dev_kfree_skb_any(pool->rx_buff[bufidx].skb);
872 		pool->rx_buff[bufidx].skb = NULL;
873 	}
874 	adapter->replenish_add_buff_failure += ind_bufp->index;
875 	atomic_add(buffers_added, &pool->available);
876 	ind_bufp->index = 0;
877 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
878 		/* Disable buffer pool replenishment and report carrier off if
879 		 * queue is closed or pending failover.
880 		 * Firmware guarantees that a signal will be sent to the
881 		 * driver, triggering a reset.
882 		 */
883 		deactivate_rx_pools(adapter);
884 		netif_carrier_off(adapter->netdev);
885 	}
886 }
887 
888 static void replenish_pools(struct ibmvnic_adapter *adapter)
889 {
890 	int i;
891 
892 	adapter->replenish_task_cycles++;
893 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
894 		if (adapter->rx_pool[i].active)
895 			replenish_rx_pool(adapter, &adapter->rx_pool[i]);
896 	}
897 
898 	netdev_dbg(adapter->netdev, "Replenished %d pools\n", i);
899 }
900 
901 static void release_stats_buffers(struct ibmvnic_adapter *adapter)
902 {
903 	kfree(adapter->tx_stats_buffers);
904 	kfree(adapter->rx_stats_buffers);
905 	adapter->tx_stats_buffers = NULL;
906 	adapter->rx_stats_buffers = NULL;
907 }
908 
909 static int init_stats_buffers(struct ibmvnic_adapter *adapter)
910 {
911 	adapter->tx_stats_buffers =
912 				kcalloc(IBMVNIC_MAX_QUEUES,
913 					sizeof(struct ibmvnic_tx_queue_stats),
914 					GFP_KERNEL);
915 	if (!adapter->tx_stats_buffers)
916 		return -ENOMEM;
917 
918 	adapter->rx_stats_buffers =
919 				kcalloc(IBMVNIC_MAX_QUEUES,
920 					sizeof(struct ibmvnic_rx_queue_stats),
921 					GFP_KERNEL);
922 	if (!adapter->rx_stats_buffers)
923 		return -ENOMEM;
924 
925 	return 0;
926 }
927 
928 static void release_stats_token(struct ibmvnic_adapter *adapter)
929 {
930 	struct device *dev = &adapter->vdev->dev;
931 
932 	if (!adapter->stats_token)
933 		return;
934 
935 	dma_unmap_single(dev, adapter->stats_token,
936 			 sizeof(struct ibmvnic_statistics),
937 			 DMA_FROM_DEVICE);
938 	adapter->stats_token = 0;
939 }
940 
941 static int init_stats_token(struct ibmvnic_adapter *adapter)
942 {
943 	struct device *dev = &adapter->vdev->dev;
944 	dma_addr_t stok;
945 	int rc;
946 
947 	stok = dma_map_single(dev, &adapter->stats,
948 			      sizeof(struct ibmvnic_statistics),
949 			      DMA_FROM_DEVICE);
950 	rc = dma_mapping_error(dev, stok);
951 	if (rc) {
952 		dev_err(dev, "Couldn't map stats buffer, rc = %d\n", rc);
953 		return rc;
954 	}
955 
956 	adapter->stats_token = stok;
957 	netdev_dbg(adapter->netdev, "Stats token initialized (%llx)\n", stok);
958 	return 0;
959 }
960 
961 /**
962  * release_rx_pools() - Release any rx pools attached to @adapter.
963  * @adapter: ibmvnic adapter
964  *
965  * Safe to call this multiple times - even if no pools are attached.
966  */
967 static void release_rx_pools(struct ibmvnic_adapter *adapter)
968 {
969 	struct ibmvnic_rx_pool *rx_pool;
970 	int i, j;
971 
972 	if (!adapter->rx_pool)
973 		return;
974 
975 	for (i = 0; i < adapter->num_active_rx_pools; i++) {
976 		rx_pool = &adapter->rx_pool[i];
977 
978 		netdev_dbg(adapter->netdev, "Releasing rx_pool[%d]\n", i);
979 
980 		kfree(rx_pool->free_map);
981 
982 		free_ltb_set(adapter, &rx_pool->ltb_set);
983 
984 		if (!rx_pool->rx_buff)
985 			continue;
986 
987 		for (j = 0; j < rx_pool->size; j++) {
988 			if (rx_pool->rx_buff[j].skb) {
989 				dev_kfree_skb_any(rx_pool->rx_buff[j].skb);
990 				rx_pool->rx_buff[j].skb = NULL;
991 			}
992 		}
993 
994 		kfree(rx_pool->rx_buff);
995 	}
996 
997 	kfree(adapter->rx_pool);
998 	adapter->rx_pool = NULL;
999 	adapter->num_active_rx_pools = 0;
1000 	adapter->prev_rx_pool_size = 0;
1001 }
1002 
1003 /**
1004  * reuse_rx_pools() - Check if the existing rx pools can be reused.
1005  * @adapter: ibmvnic adapter
1006  *
1007  * Check if the existing rx pools in the adapter can be reused. The
1008  * pools can be reused if the pool parameters (number of pools,
1009  * number of buffers in the pool and size of each buffer) have not
1010  * changed.
1011  *
1012  * NOTE: This assumes that all pools have the same number of buffers
1013  *       which is the case currently. If that changes, we must fix this.
1014  *
1015  * Return: true if the rx pools can be reused, false otherwise.
1016  */
1017 static bool reuse_rx_pools(struct ibmvnic_adapter *adapter)
1018 {
1019 	u64 old_num_pools, new_num_pools;
1020 	u64 old_pool_size, new_pool_size;
1021 	u64 old_buff_size, new_buff_size;
1022 
1023 	if (!adapter->rx_pool)
1024 		return false;
1025 
1026 	old_num_pools = adapter->num_active_rx_pools;
1027 	new_num_pools = adapter->req_rx_queues;
1028 
1029 	old_pool_size = adapter->prev_rx_pool_size;
1030 	new_pool_size = adapter->req_rx_add_entries_per_subcrq;
1031 
1032 	old_buff_size = adapter->prev_rx_buf_sz;
1033 	new_buff_size = adapter->cur_rx_buf_sz;
1034 
1035 	if (old_buff_size != new_buff_size ||
1036 	    old_num_pools != new_num_pools ||
1037 	    old_pool_size != new_pool_size)
1038 		return false;
1039 
1040 	return true;
1041 }
1042 
1043 /**
1044  * init_rx_pools(): Initialize the set of receiver pools in the adapter.
1045  * @netdev: net device associated with the vnic interface
1046  *
1047  * Initialize the set of receiver pools in the ibmvnic adapter associated
1048  * with the net_device @netdev. If possible, reuse the existing rx pools.
1049  * Otherwise free any existing pools and  allocate a new set of pools
1050  * before initializing them.
1051  *
1052  * Return: 0 on success and negative value on error.
1053  */
1054 static int init_rx_pools(struct net_device *netdev)
1055 {
1056 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1057 	struct device *dev = &adapter->vdev->dev;
1058 	struct ibmvnic_rx_pool *rx_pool;
1059 	u64 num_pools;
1060 	u64 pool_size;		/* # of buffers in one pool */
1061 	u64 buff_size;
1062 	int i, j, rc;
1063 
1064 	pool_size = adapter->req_rx_add_entries_per_subcrq;
1065 	num_pools = adapter->req_rx_queues;
1066 	buff_size = adapter->cur_rx_buf_sz;
1067 
1068 	if (reuse_rx_pools(adapter)) {
1069 		dev_dbg(dev, "Reusing rx pools\n");
1070 		goto update_ltb;
1071 	}
1072 
1073 	/* Allocate/populate the pools. */
1074 	release_rx_pools(adapter);
1075 
1076 	adapter->rx_pool = kcalloc(num_pools,
1077 				   sizeof(struct ibmvnic_rx_pool),
1078 				   GFP_KERNEL);
1079 	if (!adapter->rx_pool) {
1080 		dev_err(dev, "Failed to allocate rx pools\n");
1081 		return -ENOMEM;
1082 	}
1083 
1084 	/* Set num_active_rx_pools early. If we fail below after partial
1085 	 * allocation, release_rx_pools() will know how many to look for.
1086 	 */
1087 	adapter->num_active_rx_pools = num_pools;
1088 
1089 	for (i = 0; i < num_pools; i++) {
1090 		rx_pool = &adapter->rx_pool[i];
1091 
1092 		netdev_dbg(adapter->netdev,
1093 			   "Initializing rx_pool[%d], %lld buffs, %lld bytes each\n",
1094 			   i, pool_size, buff_size);
1095 
1096 		rx_pool->size = pool_size;
1097 		rx_pool->index = i;
1098 		rx_pool->buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1099 
1100 		rx_pool->free_map = kcalloc(rx_pool->size, sizeof(int),
1101 					    GFP_KERNEL);
1102 		if (!rx_pool->free_map) {
1103 			dev_err(dev, "Couldn't alloc free_map %d\n", i);
1104 			rc = -ENOMEM;
1105 			goto out_release;
1106 		}
1107 
1108 		rx_pool->rx_buff = kcalloc(rx_pool->size,
1109 					   sizeof(struct ibmvnic_rx_buff),
1110 					   GFP_KERNEL);
1111 		if (!rx_pool->rx_buff) {
1112 			dev_err(dev, "Couldn't alloc rx buffers\n");
1113 			rc = -ENOMEM;
1114 			goto out_release;
1115 		}
1116 	}
1117 
1118 	adapter->prev_rx_pool_size = pool_size;
1119 	adapter->prev_rx_buf_sz = adapter->cur_rx_buf_sz;
1120 
1121 update_ltb:
1122 	for (i = 0; i < num_pools; i++) {
1123 		rx_pool = &adapter->rx_pool[i];
1124 		dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
1125 			i, rx_pool->size, rx_pool->buff_size);
1126 
1127 		rc = alloc_ltb_set(adapter, &rx_pool->ltb_set,
1128 				   rx_pool->size, rx_pool->buff_size);
1129 		if (rc)
1130 			goto out;
1131 
1132 		for (j = 0; j < rx_pool->size; ++j) {
1133 			struct ibmvnic_rx_buff *rx_buff;
1134 
1135 			rx_pool->free_map[j] = j;
1136 
1137 			/* NOTE: Don't clear rx_buff->skb here - will leak
1138 			 * memory! replenish_rx_pool() will reuse skbs or
1139 			 * allocate as necessary.
1140 			 */
1141 			rx_buff = &rx_pool->rx_buff[j];
1142 			rx_buff->dma = 0;
1143 			rx_buff->data = 0;
1144 			rx_buff->size = 0;
1145 			rx_buff->pool_index = 0;
1146 		}
1147 
1148 		/* Mark pool "empty" so replenish_rx_pools() will
1149 		 * update the LTB info for each buffer
1150 		 */
1151 		atomic_set(&rx_pool->available, 0);
1152 		rx_pool->next_alloc = 0;
1153 		rx_pool->next_free = 0;
1154 		/* replenish_rx_pool() may have called deactivate_rx_pools()
1155 		 * on failover. Ensure pool is active now.
1156 		 */
1157 		rx_pool->active = 1;
1158 	}
1159 	return 0;
1160 out_release:
1161 	release_rx_pools(adapter);
1162 out:
1163 	/* We failed to allocate one or more LTBs or map them on the VIOS.
1164 	 * Hold onto the pools and any LTBs that we did allocate/map.
1165 	 */
1166 	return rc;
1167 }
1168 
1169 static void release_vpd_data(struct ibmvnic_adapter *adapter)
1170 {
1171 	if (!adapter->vpd)
1172 		return;
1173 
1174 	kfree(adapter->vpd->buff);
1175 	kfree(adapter->vpd);
1176 
1177 	adapter->vpd = NULL;
1178 }
1179 
1180 static void release_one_tx_pool(struct ibmvnic_adapter *adapter,
1181 				struct ibmvnic_tx_pool *tx_pool)
1182 {
1183 	kfree(tx_pool->tx_buff);
1184 	kfree(tx_pool->free_map);
1185 	free_ltb_set(adapter, &tx_pool->ltb_set);
1186 }
1187 
1188 /**
1189  * release_tx_pools() - Release any tx pools attached to @adapter.
1190  * @adapter: ibmvnic adapter
1191  *
1192  * Safe to call this multiple times - even if no pools are attached.
1193  */
1194 static void release_tx_pools(struct ibmvnic_adapter *adapter)
1195 {
1196 	int i;
1197 
1198 	/* init_tx_pools() ensures that ->tx_pool and ->tso_pool are
1199 	 * both NULL or both non-NULL. So we only need to check one.
1200 	 */
1201 	if (!adapter->tx_pool)
1202 		return;
1203 
1204 	for (i = 0; i < adapter->num_active_tx_pools; i++) {
1205 		release_one_tx_pool(adapter, &adapter->tx_pool[i]);
1206 		release_one_tx_pool(adapter, &adapter->tso_pool[i]);
1207 	}
1208 
1209 	kfree(adapter->tx_pool);
1210 	adapter->tx_pool = NULL;
1211 	kfree(adapter->tso_pool);
1212 	adapter->tso_pool = NULL;
1213 	adapter->num_active_tx_pools = 0;
1214 	adapter->prev_tx_pool_size = 0;
1215 }
1216 
1217 static int init_one_tx_pool(struct net_device *netdev,
1218 			    struct ibmvnic_tx_pool *tx_pool,
1219 			    int pool_size, int buf_size)
1220 {
1221 	int i;
1222 
1223 	tx_pool->tx_buff = kcalloc(pool_size,
1224 				   sizeof(struct ibmvnic_tx_buff),
1225 				   GFP_KERNEL);
1226 	if (!tx_pool->tx_buff)
1227 		return -ENOMEM;
1228 
1229 	tx_pool->free_map = kcalloc(pool_size, sizeof(int), GFP_KERNEL);
1230 	if (!tx_pool->free_map) {
1231 		kfree(tx_pool->tx_buff);
1232 		tx_pool->tx_buff = NULL;
1233 		return -ENOMEM;
1234 	}
1235 
1236 	for (i = 0; i < pool_size; i++)
1237 		tx_pool->free_map[i] = i;
1238 
1239 	tx_pool->consumer_index = 0;
1240 	tx_pool->producer_index = 0;
1241 	tx_pool->num_buffers = pool_size;
1242 	tx_pool->buf_size = buf_size;
1243 
1244 	return 0;
1245 }
1246 
1247 /**
1248  * reuse_tx_pools() - Check if the existing tx pools can be reused.
1249  * @adapter: ibmvnic adapter
1250  *
1251  * Check if the existing tx pools in the adapter can be reused. The
1252  * pools can be reused if the pool parameters (number of pools,
1253  * number of buffers in the pool and mtu) have not changed.
1254  *
1255  * NOTE: This assumes that all pools have the same number of buffers
1256  *       which is the case currently. If that changes, we must fix this.
1257  *
1258  * Return: true if the tx pools can be reused, false otherwise.
1259  */
1260 static bool reuse_tx_pools(struct ibmvnic_adapter *adapter)
1261 {
1262 	u64 old_num_pools, new_num_pools;
1263 	u64 old_pool_size, new_pool_size;
1264 	u64 old_mtu, new_mtu;
1265 
1266 	if (!adapter->tx_pool)
1267 		return false;
1268 
1269 	old_num_pools = adapter->num_active_tx_pools;
1270 	new_num_pools = adapter->num_active_tx_scrqs;
1271 	old_pool_size = adapter->prev_tx_pool_size;
1272 	new_pool_size = adapter->req_tx_entries_per_subcrq;
1273 	old_mtu = adapter->prev_mtu;
1274 	new_mtu = adapter->req_mtu;
1275 
1276 	if (old_mtu != new_mtu ||
1277 	    old_num_pools != new_num_pools ||
1278 	    old_pool_size != new_pool_size)
1279 		return false;
1280 
1281 	return true;
1282 }
1283 
1284 /**
1285  * init_tx_pools(): Initialize the set of transmit pools in the adapter.
1286  * @netdev: net device associated with the vnic interface
1287  *
1288  * Initialize the set of transmit pools in the ibmvnic adapter associated
1289  * with the net_device @netdev. If possible, reuse the existing tx pools.
1290  * Otherwise free any existing pools and  allocate a new set of pools
1291  * before initializing them.
1292  *
1293  * Return: 0 on success and negative value on error.
1294  */
1295 static int init_tx_pools(struct net_device *netdev)
1296 {
1297 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1298 	struct device *dev = &adapter->vdev->dev;
1299 	int num_pools;
1300 	u64 pool_size;		/* # of buffers in pool */
1301 	u64 buff_size;
1302 	int i, j, rc;
1303 
1304 	num_pools = adapter->req_tx_queues;
1305 
1306 	/* We must notify the VIOS about the LTB on all resets - but we only
1307 	 * need to alloc/populate pools if either the number of buffers or
1308 	 * size of each buffer in the pool has changed.
1309 	 */
1310 	if (reuse_tx_pools(adapter)) {
1311 		netdev_dbg(netdev, "Reusing tx pools\n");
1312 		goto update_ltb;
1313 	}
1314 
1315 	/* Allocate/populate the pools. */
1316 	release_tx_pools(adapter);
1317 
1318 	pool_size = adapter->req_tx_entries_per_subcrq;
1319 	num_pools = adapter->num_active_tx_scrqs;
1320 
1321 	adapter->tx_pool = kcalloc(num_pools,
1322 				   sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1323 	if (!adapter->tx_pool)
1324 		return -ENOMEM;
1325 
1326 	adapter->tso_pool = kcalloc(num_pools,
1327 				    sizeof(struct ibmvnic_tx_pool), GFP_KERNEL);
1328 	/* To simplify release_tx_pools() ensure that ->tx_pool and
1329 	 * ->tso_pool are either both NULL or both non-NULL.
1330 	 */
1331 	if (!adapter->tso_pool) {
1332 		kfree(adapter->tx_pool);
1333 		adapter->tx_pool = NULL;
1334 		return -ENOMEM;
1335 	}
1336 
1337 	/* Set num_active_tx_pools early. If we fail below after partial
1338 	 * allocation, release_tx_pools() will know how many to look for.
1339 	 */
1340 	adapter->num_active_tx_pools = num_pools;
1341 
1342 	buff_size = adapter->req_mtu + VLAN_HLEN;
1343 	buff_size = ALIGN(buff_size, L1_CACHE_BYTES);
1344 
1345 	for (i = 0; i < num_pools; i++) {
1346 		dev_dbg(dev, "Init tx pool %d [%llu, %llu]\n",
1347 			i, adapter->req_tx_entries_per_subcrq, buff_size);
1348 
1349 		rc = init_one_tx_pool(netdev, &adapter->tx_pool[i],
1350 				      pool_size, buff_size);
1351 		if (rc)
1352 			goto out_release;
1353 
1354 		rc = init_one_tx_pool(netdev, &adapter->tso_pool[i],
1355 				      IBMVNIC_TSO_BUFS,
1356 				      IBMVNIC_TSO_BUF_SZ);
1357 		if (rc)
1358 			goto out_release;
1359 	}
1360 
1361 	adapter->prev_tx_pool_size = pool_size;
1362 	adapter->prev_mtu = adapter->req_mtu;
1363 
1364 update_ltb:
1365 	/* NOTE: All tx_pools have the same number of buffers (which is
1366 	 *       same as pool_size). All tso_pools have IBMVNIC_TSO_BUFS
1367 	 *       buffers (see calls init_one_tx_pool() for these).
1368 	 *       For consistency, we use tx_pool->num_buffers and
1369 	 *       tso_pool->num_buffers below.
1370 	 */
1371 	rc = -1;
1372 	for (i = 0; i < num_pools; i++) {
1373 		struct ibmvnic_tx_pool *tso_pool;
1374 		struct ibmvnic_tx_pool *tx_pool;
1375 
1376 		tx_pool = &adapter->tx_pool[i];
1377 
1378 		dev_dbg(dev, "Updating LTB for tx pool %d [%d, %d]\n",
1379 			i, tx_pool->num_buffers, tx_pool->buf_size);
1380 
1381 		rc = alloc_ltb_set(adapter, &tx_pool->ltb_set,
1382 				   tx_pool->num_buffers, tx_pool->buf_size);
1383 		if (rc)
1384 			goto out;
1385 
1386 		tx_pool->consumer_index = 0;
1387 		tx_pool->producer_index = 0;
1388 
1389 		for (j = 0; j < tx_pool->num_buffers; j++)
1390 			tx_pool->free_map[j] = j;
1391 
1392 		tso_pool = &adapter->tso_pool[i];
1393 
1394 		dev_dbg(dev, "Updating LTB for tso pool %d [%d, %d]\n",
1395 			i, tso_pool->num_buffers, tso_pool->buf_size);
1396 
1397 		rc = alloc_ltb_set(adapter, &tso_pool->ltb_set,
1398 				   tso_pool->num_buffers, tso_pool->buf_size);
1399 		if (rc)
1400 			goto out;
1401 
1402 		tso_pool->consumer_index = 0;
1403 		tso_pool->producer_index = 0;
1404 
1405 		for (j = 0; j < tso_pool->num_buffers; j++)
1406 			tso_pool->free_map[j] = j;
1407 	}
1408 
1409 	return 0;
1410 out_release:
1411 	release_tx_pools(adapter);
1412 out:
1413 	/* We failed to allocate one or more LTBs or map them on the VIOS.
1414 	 * Hold onto the pools and any LTBs that we did allocate/map.
1415 	 */
1416 	return rc;
1417 }
1418 
1419 static void ibmvnic_napi_enable(struct ibmvnic_adapter *adapter)
1420 {
1421 	int i;
1422 
1423 	if (adapter->napi_enabled)
1424 		return;
1425 
1426 	for (i = 0; i < adapter->req_rx_queues; i++)
1427 		napi_enable(&adapter->napi[i]);
1428 
1429 	adapter->napi_enabled = true;
1430 }
1431 
1432 static void ibmvnic_napi_disable(struct ibmvnic_adapter *adapter)
1433 {
1434 	int i;
1435 
1436 	if (!adapter->napi_enabled)
1437 		return;
1438 
1439 	for (i = 0; i < adapter->req_rx_queues; i++) {
1440 		netdev_dbg(adapter->netdev, "Disabling napi[%d]\n", i);
1441 		napi_disable(&adapter->napi[i]);
1442 	}
1443 
1444 	adapter->napi_enabled = false;
1445 }
1446 
1447 static int init_napi(struct ibmvnic_adapter *adapter)
1448 {
1449 	int i;
1450 
1451 	adapter->napi = kcalloc(adapter->req_rx_queues,
1452 				sizeof(struct napi_struct), GFP_KERNEL);
1453 	if (!adapter->napi)
1454 		return -ENOMEM;
1455 
1456 	for (i = 0; i < adapter->req_rx_queues; i++) {
1457 		netdev_dbg(adapter->netdev, "Adding napi[%d]\n", i);
1458 		netif_napi_add(adapter->netdev, &adapter->napi[i],
1459 			       ibmvnic_poll);
1460 	}
1461 
1462 	adapter->num_active_rx_napi = adapter->req_rx_queues;
1463 	return 0;
1464 }
1465 
1466 static void release_napi(struct ibmvnic_adapter *adapter)
1467 {
1468 	int i;
1469 
1470 	if (!adapter->napi)
1471 		return;
1472 
1473 	for (i = 0; i < adapter->num_active_rx_napi; i++) {
1474 		netdev_dbg(adapter->netdev, "Releasing napi[%d]\n", i);
1475 		netif_napi_del(&adapter->napi[i]);
1476 	}
1477 
1478 	kfree(adapter->napi);
1479 	adapter->napi = NULL;
1480 	adapter->num_active_rx_napi = 0;
1481 	adapter->napi_enabled = false;
1482 }
1483 
1484 static const char *adapter_state_to_string(enum vnic_state state)
1485 {
1486 	switch (state) {
1487 	case VNIC_PROBING:
1488 		return "PROBING";
1489 	case VNIC_PROBED:
1490 		return "PROBED";
1491 	case VNIC_OPENING:
1492 		return "OPENING";
1493 	case VNIC_OPEN:
1494 		return "OPEN";
1495 	case VNIC_CLOSING:
1496 		return "CLOSING";
1497 	case VNIC_CLOSED:
1498 		return "CLOSED";
1499 	case VNIC_REMOVING:
1500 		return "REMOVING";
1501 	case VNIC_REMOVED:
1502 		return "REMOVED";
1503 	case VNIC_DOWN:
1504 		return "DOWN";
1505 	}
1506 	return "UNKNOWN";
1507 }
1508 
1509 static int ibmvnic_login(struct net_device *netdev)
1510 {
1511 	unsigned long flags, timeout = msecs_to_jiffies(20000);
1512 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1513 	int retry_count = 0;
1514 	int retries = 10;
1515 	bool retry;
1516 	int rc;
1517 
1518 	do {
1519 		retry = false;
1520 		if (retry_count > retries) {
1521 			netdev_warn(netdev, "Login attempts exceeded\n");
1522 			return -EACCES;
1523 		}
1524 
1525 		adapter->init_done_rc = 0;
1526 		reinit_completion(&adapter->init_done);
1527 		rc = send_login(adapter);
1528 		if (rc)
1529 			return rc;
1530 
1531 		if (!wait_for_completion_timeout(&adapter->init_done,
1532 						 timeout)) {
1533 			netdev_warn(netdev, "Login timed out\n");
1534 			adapter->login_pending = false;
1535 			goto partial_reset;
1536 		}
1537 
1538 		if (adapter->init_done_rc == ABORTED) {
1539 			netdev_warn(netdev, "Login aborted, retrying...\n");
1540 			retry = true;
1541 			adapter->init_done_rc = 0;
1542 			retry_count++;
1543 			/* FW or device may be busy, so
1544 			 * wait a bit before retrying login
1545 			 */
1546 			msleep(500);
1547 		} else if (adapter->init_done_rc == PARTIALSUCCESS) {
1548 			retry_count++;
1549 			release_sub_crqs(adapter, 1);
1550 
1551 			retry = true;
1552 			netdev_dbg(netdev,
1553 				   "Received partial success, retrying...\n");
1554 			adapter->init_done_rc = 0;
1555 			reinit_completion(&adapter->init_done);
1556 			send_query_cap(adapter);
1557 			if (!wait_for_completion_timeout(&adapter->init_done,
1558 							 timeout)) {
1559 				netdev_warn(netdev,
1560 					    "Capabilities query timed out\n");
1561 				return -ETIMEDOUT;
1562 			}
1563 
1564 			rc = init_sub_crqs(adapter);
1565 			if (rc) {
1566 				netdev_warn(netdev,
1567 					    "SCRQ initialization failed\n");
1568 				return rc;
1569 			}
1570 
1571 			rc = init_sub_crq_irqs(adapter);
1572 			if (rc) {
1573 				netdev_warn(netdev,
1574 					    "SCRQ irq initialization failed\n");
1575 				return rc;
1576 			}
1577 		/* Default/timeout error handling, reset and start fresh */
1578 		} else if (adapter->init_done_rc) {
1579 			netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n",
1580 				    adapter->init_done_rc);
1581 
1582 partial_reset:
1583 			/* adapter login failed, so free any CRQs or sub-CRQs
1584 			 * and register again before attempting to login again.
1585 			 * If we don't do this then the VIOS may think that
1586 			 * we are already logged in and reject any subsequent
1587 			 * attempts
1588 			 */
1589 			netdev_warn(netdev,
1590 				    "Freeing and re-registering CRQs before attempting to login again\n");
1591 			retry = true;
1592 			adapter->init_done_rc = 0;
1593 			release_sub_crqs(adapter, true);
1594 			/* Much of this is similar logic as ibmvnic_probe(),
1595 			 * we are essentially re-initializing communication
1596 			 * with the server. We really should not run any
1597 			 * resets/failovers here because this is already a form
1598 			 * of reset and we do not want parallel resets occurring
1599 			 */
1600 			do {
1601 				reinit_init_done(adapter);
1602 				/* Clear any failovers we got in the previous
1603 				 * pass since we are re-initializing the CRQ
1604 				 */
1605 				adapter->failover_pending = false;
1606 				release_crq_queue(adapter);
1607 				/* If we don't sleep here then we risk an
1608 				 * unnecessary failover event from the VIOS.
1609 				 * This is a known VIOS issue caused by a vnic
1610 				 * device freeing and registering a CRQ too
1611 				 * quickly.
1612 				 */
1613 				msleep(1500);
1614 				/* Avoid any resets, since we are currently
1615 				 * resetting.
1616 				 */
1617 				spin_lock_irqsave(&adapter->rwi_lock, flags);
1618 				flush_reset_queue(adapter);
1619 				spin_unlock_irqrestore(&adapter->rwi_lock,
1620 						       flags);
1621 
1622 				rc = init_crq_queue(adapter);
1623 				if (rc) {
1624 					netdev_err(netdev, "login recovery: init CRQ failed %d\n",
1625 						   rc);
1626 					return -EIO;
1627 				}
1628 
1629 				rc = ibmvnic_reset_init(adapter, false);
1630 				if (rc)
1631 					netdev_err(netdev, "login recovery: Reset init failed %d\n",
1632 						   rc);
1633 				/* IBMVNIC_CRQ_INIT will return EAGAIN if it
1634 				 * fails, since ibmvnic_reset_init will free
1635 				 * irq's in failure, we won't be able to receive
1636 				 * new CRQs so we need to keep trying. probe()
1637 				 * handles this similarly.
1638 				 */
1639 			} while (rc == -EAGAIN && retry_count++ < retries);
1640 		}
1641 	} while (retry);
1642 
1643 	__ibmvnic_set_mac(netdev, adapter->mac_addr);
1644 
1645 	netdev_dbg(netdev, "[S:%s] Login succeeded\n", adapter_state_to_string(adapter->state));
1646 	return 0;
1647 }
1648 
1649 static void release_login_buffer(struct ibmvnic_adapter *adapter)
1650 {
1651 	if (!adapter->login_buf)
1652 		return;
1653 
1654 	dma_unmap_single(&adapter->vdev->dev, adapter->login_buf_token,
1655 			 adapter->login_buf_sz, DMA_TO_DEVICE);
1656 	kfree(adapter->login_buf);
1657 	adapter->login_buf = NULL;
1658 }
1659 
1660 static void release_login_rsp_buffer(struct ibmvnic_adapter *adapter)
1661 {
1662 	if (!adapter->login_rsp_buf)
1663 		return;
1664 
1665 	dma_unmap_single(&adapter->vdev->dev, adapter->login_rsp_buf_token,
1666 			 adapter->login_rsp_buf_sz, DMA_FROM_DEVICE);
1667 	kfree(adapter->login_rsp_buf);
1668 	adapter->login_rsp_buf = NULL;
1669 }
1670 
1671 static void release_resources(struct ibmvnic_adapter *adapter)
1672 {
1673 	release_vpd_data(adapter);
1674 
1675 	release_napi(adapter);
1676 	release_login_buffer(adapter);
1677 	release_login_rsp_buffer(adapter);
1678 }
1679 
1680 static int set_link_state(struct ibmvnic_adapter *adapter, u8 link_state)
1681 {
1682 	struct net_device *netdev = adapter->netdev;
1683 	unsigned long timeout = msecs_to_jiffies(20000);
1684 	union ibmvnic_crq crq;
1685 	bool resend;
1686 	int rc;
1687 
1688 	netdev_dbg(netdev, "setting link state %d\n", link_state);
1689 
1690 	memset(&crq, 0, sizeof(crq));
1691 	crq.logical_link_state.first = IBMVNIC_CRQ_CMD;
1692 	crq.logical_link_state.cmd = LOGICAL_LINK_STATE;
1693 	crq.logical_link_state.link_state = link_state;
1694 
1695 	do {
1696 		resend = false;
1697 
1698 		reinit_completion(&adapter->init_done);
1699 		rc = ibmvnic_send_crq(adapter, &crq);
1700 		if (rc) {
1701 			netdev_err(netdev, "Failed to set link state\n");
1702 			return rc;
1703 		}
1704 
1705 		if (!wait_for_completion_timeout(&adapter->init_done,
1706 						 timeout)) {
1707 			netdev_err(netdev, "timeout setting link state\n");
1708 			return -ETIMEDOUT;
1709 		}
1710 
1711 		if (adapter->init_done_rc == PARTIALSUCCESS) {
1712 			/* Partuial success, delay and re-send */
1713 			mdelay(1000);
1714 			resend = true;
1715 		} else if (adapter->init_done_rc) {
1716 			netdev_warn(netdev, "Unable to set link state, rc=%d\n",
1717 				    adapter->init_done_rc);
1718 			return adapter->init_done_rc;
1719 		}
1720 	} while (resend);
1721 
1722 	return 0;
1723 }
1724 
1725 static int set_real_num_queues(struct net_device *netdev)
1726 {
1727 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1728 	int rc;
1729 
1730 	netdev_dbg(netdev, "Setting real tx/rx queues (%llx/%llx)\n",
1731 		   adapter->req_tx_queues, adapter->req_rx_queues);
1732 
1733 	rc = netif_set_real_num_tx_queues(netdev, adapter->req_tx_queues);
1734 	if (rc) {
1735 		netdev_err(netdev, "failed to set the number of tx queues\n");
1736 		return rc;
1737 	}
1738 
1739 	rc = netif_set_real_num_rx_queues(netdev, adapter->req_rx_queues);
1740 	if (rc)
1741 		netdev_err(netdev, "failed to set the number of rx queues\n");
1742 
1743 	return rc;
1744 }
1745 
1746 static int ibmvnic_get_vpd(struct ibmvnic_adapter *adapter)
1747 {
1748 	struct device *dev = &adapter->vdev->dev;
1749 	union ibmvnic_crq crq;
1750 	int len = 0;
1751 	int rc;
1752 
1753 	if (adapter->vpd->buff)
1754 		len = adapter->vpd->len;
1755 
1756 	mutex_lock(&adapter->fw_lock);
1757 	adapter->fw_done_rc = 0;
1758 	reinit_completion(&adapter->fw_done);
1759 
1760 	crq.get_vpd_size.first = IBMVNIC_CRQ_CMD;
1761 	crq.get_vpd_size.cmd = GET_VPD_SIZE;
1762 	rc = ibmvnic_send_crq(adapter, &crq);
1763 	if (rc) {
1764 		mutex_unlock(&adapter->fw_lock);
1765 		return rc;
1766 	}
1767 
1768 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1769 	if (rc) {
1770 		dev_err(dev, "Could not retrieve VPD size, rc = %d\n", rc);
1771 		mutex_unlock(&adapter->fw_lock);
1772 		return rc;
1773 	}
1774 	mutex_unlock(&adapter->fw_lock);
1775 
1776 	if (!adapter->vpd->len)
1777 		return -ENODATA;
1778 
1779 	if (!adapter->vpd->buff)
1780 		adapter->vpd->buff = kzalloc(adapter->vpd->len, GFP_KERNEL);
1781 	else if (adapter->vpd->len != len)
1782 		adapter->vpd->buff =
1783 			krealloc(adapter->vpd->buff,
1784 				 adapter->vpd->len, GFP_KERNEL);
1785 
1786 	if (!adapter->vpd->buff) {
1787 		dev_err(dev, "Could allocate VPD buffer\n");
1788 		return -ENOMEM;
1789 	}
1790 
1791 	adapter->vpd->dma_addr =
1792 		dma_map_single(dev, adapter->vpd->buff, adapter->vpd->len,
1793 			       DMA_FROM_DEVICE);
1794 	if (dma_mapping_error(dev, adapter->vpd->dma_addr)) {
1795 		dev_err(dev, "Could not map VPD buffer\n");
1796 		kfree(adapter->vpd->buff);
1797 		adapter->vpd->buff = NULL;
1798 		return -ENOMEM;
1799 	}
1800 
1801 	mutex_lock(&adapter->fw_lock);
1802 	adapter->fw_done_rc = 0;
1803 	reinit_completion(&adapter->fw_done);
1804 
1805 	crq.get_vpd.first = IBMVNIC_CRQ_CMD;
1806 	crq.get_vpd.cmd = GET_VPD;
1807 	crq.get_vpd.ioba = cpu_to_be32(adapter->vpd->dma_addr);
1808 	crq.get_vpd.len = cpu_to_be32((u32)adapter->vpd->len);
1809 	rc = ibmvnic_send_crq(adapter, &crq);
1810 	if (rc) {
1811 		kfree(adapter->vpd->buff);
1812 		adapter->vpd->buff = NULL;
1813 		mutex_unlock(&adapter->fw_lock);
1814 		return rc;
1815 	}
1816 
1817 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
1818 	if (rc) {
1819 		dev_err(dev, "Unable to retrieve VPD, rc = %d\n", rc);
1820 		kfree(adapter->vpd->buff);
1821 		adapter->vpd->buff = NULL;
1822 		mutex_unlock(&adapter->fw_lock);
1823 		return rc;
1824 	}
1825 
1826 	mutex_unlock(&adapter->fw_lock);
1827 	return 0;
1828 }
1829 
1830 static int init_resources(struct ibmvnic_adapter *adapter)
1831 {
1832 	struct net_device *netdev = adapter->netdev;
1833 	int rc;
1834 
1835 	rc = set_real_num_queues(netdev);
1836 	if (rc)
1837 		return rc;
1838 
1839 	adapter->vpd = kzalloc(sizeof(*adapter->vpd), GFP_KERNEL);
1840 	if (!adapter->vpd)
1841 		return -ENOMEM;
1842 
1843 	/* Vital Product Data (VPD) */
1844 	rc = ibmvnic_get_vpd(adapter);
1845 	if (rc) {
1846 		netdev_err(netdev, "failed to initialize Vital Product Data (VPD)\n");
1847 		return rc;
1848 	}
1849 
1850 	rc = init_napi(adapter);
1851 	if (rc)
1852 		return rc;
1853 
1854 	send_query_map(adapter);
1855 
1856 	rc = init_rx_pools(netdev);
1857 	if (rc)
1858 		return rc;
1859 
1860 	rc = init_tx_pools(netdev);
1861 	return rc;
1862 }
1863 
1864 static int __ibmvnic_open(struct net_device *netdev)
1865 {
1866 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1867 	enum vnic_state prev_state = adapter->state;
1868 	int i, rc;
1869 
1870 	adapter->state = VNIC_OPENING;
1871 	replenish_pools(adapter);
1872 	ibmvnic_napi_enable(adapter);
1873 
1874 	/* We're ready to receive frames, enable the sub-crq interrupts and
1875 	 * set the logical link state to up
1876 	 */
1877 	for (i = 0; i < adapter->req_rx_queues; i++) {
1878 		netdev_dbg(netdev, "Enabling rx_scrq[%d] irq\n", i);
1879 		if (prev_state == VNIC_CLOSED)
1880 			enable_irq(adapter->rx_scrq[i]->irq);
1881 		enable_scrq_irq(adapter, adapter->rx_scrq[i]);
1882 	}
1883 
1884 	for (i = 0; i < adapter->req_tx_queues; i++) {
1885 		netdev_dbg(netdev, "Enabling tx_scrq[%d] irq\n", i);
1886 		if (prev_state == VNIC_CLOSED)
1887 			enable_irq(adapter->tx_scrq[i]->irq);
1888 		enable_scrq_irq(adapter, adapter->tx_scrq[i]);
1889 		/* netdev_tx_reset_queue will reset dql stats. During NON_FATAL
1890 		 * resets, don't reset the stats because there could be batched
1891 		 * skb's waiting to be sent. If we reset dql stats, we risk
1892 		 * num_completed being greater than num_queued. This will cause
1893 		 * a BUG_ON in dql_completed().
1894 		 */
1895 		if (adapter->reset_reason != VNIC_RESET_NON_FATAL)
1896 			netdev_tx_reset_queue(netdev_get_tx_queue(netdev, i));
1897 	}
1898 
1899 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_UP);
1900 	if (rc) {
1901 		ibmvnic_napi_disable(adapter);
1902 		ibmvnic_disable_irqs(adapter);
1903 		return rc;
1904 	}
1905 
1906 	adapter->tx_queues_active = true;
1907 
1908 	/* Since queues were stopped until now, there shouldn't be any
1909 	 * one in ibmvnic_complete_tx() or ibmvnic_xmit() so maybe we
1910 	 * don't need the synchronize_rcu()? Leaving it for consistency
1911 	 * with setting ->tx_queues_active = false.
1912 	 */
1913 	synchronize_rcu();
1914 
1915 	netif_tx_start_all_queues(netdev);
1916 
1917 	if (prev_state == VNIC_CLOSED) {
1918 		for (i = 0; i < adapter->req_rx_queues; i++)
1919 			napi_schedule(&adapter->napi[i]);
1920 	}
1921 
1922 	adapter->state = VNIC_OPEN;
1923 	return rc;
1924 }
1925 
1926 static int ibmvnic_open(struct net_device *netdev)
1927 {
1928 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
1929 	int rc;
1930 
1931 	ASSERT_RTNL();
1932 
1933 	/* If device failover is pending or we are about to reset, just set
1934 	 * device state and return. Device operation will be handled by reset
1935 	 * routine.
1936 	 *
1937 	 * It should be safe to overwrite the adapter->state here. Since
1938 	 * we hold the rtnl, either the reset has not actually started or
1939 	 * the rtnl got dropped during the set_link_state() in do_reset().
1940 	 * In the former case, no one else is changing the state (again we
1941 	 * have the rtnl) and in the latter case, do_reset() will detect and
1942 	 * honor our setting below.
1943 	 */
1944 	if (adapter->failover_pending || (test_bit(0, &adapter->resetting))) {
1945 		netdev_dbg(netdev, "[S:%s FOP:%d] Resetting, deferring open\n",
1946 			   adapter_state_to_string(adapter->state),
1947 			   adapter->failover_pending);
1948 		adapter->state = VNIC_OPEN;
1949 		rc = 0;
1950 		goto out;
1951 	}
1952 
1953 	if (adapter->state != VNIC_CLOSED) {
1954 		rc = ibmvnic_login(netdev);
1955 		if (rc)
1956 			goto out;
1957 
1958 		rc = init_resources(adapter);
1959 		if (rc) {
1960 			netdev_err(netdev, "failed to initialize resources\n");
1961 			goto out;
1962 		}
1963 	}
1964 
1965 	rc = __ibmvnic_open(netdev);
1966 
1967 out:
1968 	/* If open failed and there is a pending failover or in-progress reset,
1969 	 * set device state and return. Device operation will be handled by
1970 	 * reset routine. See also comments above regarding rtnl.
1971 	 */
1972 	if (rc &&
1973 	    (adapter->failover_pending || (test_bit(0, &adapter->resetting)))) {
1974 		adapter->state = VNIC_OPEN;
1975 		rc = 0;
1976 	}
1977 
1978 	if (rc) {
1979 		release_resources(adapter);
1980 		release_rx_pools(adapter);
1981 		release_tx_pools(adapter);
1982 	}
1983 
1984 	return rc;
1985 }
1986 
1987 static void clean_rx_pools(struct ibmvnic_adapter *adapter)
1988 {
1989 	struct ibmvnic_rx_pool *rx_pool;
1990 	struct ibmvnic_rx_buff *rx_buff;
1991 	u64 rx_entries;
1992 	int rx_scrqs;
1993 	int i, j;
1994 
1995 	if (!adapter->rx_pool)
1996 		return;
1997 
1998 	rx_scrqs = adapter->num_active_rx_pools;
1999 	rx_entries = adapter->req_rx_add_entries_per_subcrq;
2000 
2001 	/* Free any remaining skbs in the rx buffer pools */
2002 	for (i = 0; i < rx_scrqs; i++) {
2003 		rx_pool = &adapter->rx_pool[i];
2004 		if (!rx_pool || !rx_pool->rx_buff)
2005 			continue;
2006 
2007 		netdev_dbg(adapter->netdev, "Cleaning rx_pool[%d]\n", i);
2008 		for (j = 0; j < rx_entries; j++) {
2009 			rx_buff = &rx_pool->rx_buff[j];
2010 			if (rx_buff && rx_buff->skb) {
2011 				dev_kfree_skb_any(rx_buff->skb);
2012 				rx_buff->skb = NULL;
2013 			}
2014 		}
2015 	}
2016 }
2017 
2018 static void clean_one_tx_pool(struct ibmvnic_adapter *adapter,
2019 			      struct ibmvnic_tx_pool *tx_pool)
2020 {
2021 	struct ibmvnic_tx_buff *tx_buff;
2022 	u64 tx_entries;
2023 	int i;
2024 
2025 	if (!tx_pool || !tx_pool->tx_buff)
2026 		return;
2027 
2028 	tx_entries = tx_pool->num_buffers;
2029 
2030 	for (i = 0; i < tx_entries; i++) {
2031 		tx_buff = &tx_pool->tx_buff[i];
2032 		if (tx_buff && tx_buff->skb) {
2033 			dev_kfree_skb_any(tx_buff->skb);
2034 			tx_buff->skb = NULL;
2035 		}
2036 	}
2037 }
2038 
2039 static void clean_tx_pools(struct ibmvnic_adapter *adapter)
2040 {
2041 	int tx_scrqs;
2042 	int i;
2043 
2044 	if (!adapter->tx_pool || !adapter->tso_pool)
2045 		return;
2046 
2047 	tx_scrqs = adapter->num_active_tx_pools;
2048 
2049 	/* Free any remaining skbs in the tx buffer pools */
2050 	for (i = 0; i < tx_scrqs; i++) {
2051 		netdev_dbg(adapter->netdev, "Cleaning tx_pool[%d]\n", i);
2052 		clean_one_tx_pool(adapter, &adapter->tx_pool[i]);
2053 		clean_one_tx_pool(adapter, &adapter->tso_pool[i]);
2054 	}
2055 }
2056 
2057 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter)
2058 {
2059 	struct net_device *netdev = adapter->netdev;
2060 	int i;
2061 
2062 	if (adapter->tx_scrq) {
2063 		for (i = 0; i < adapter->req_tx_queues; i++)
2064 			if (adapter->tx_scrq[i]->irq) {
2065 				netdev_dbg(netdev,
2066 					   "Disabling tx_scrq[%d] irq\n", i);
2067 				disable_scrq_irq(adapter, adapter->tx_scrq[i]);
2068 				disable_irq(adapter->tx_scrq[i]->irq);
2069 			}
2070 	}
2071 
2072 	if (adapter->rx_scrq) {
2073 		for (i = 0; i < adapter->req_rx_queues; i++) {
2074 			if (adapter->rx_scrq[i]->irq) {
2075 				netdev_dbg(netdev,
2076 					   "Disabling rx_scrq[%d] irq\n", i);
2077 				disable_scrq_irq(adapter, adapter->rx_scrq[i]);
2078 				disable_irq(adapter->rx_scrq[i]->irq);
2079 			}
2080 		}
2081 	}
2082 }
2083 
2084 static void ibmvnic_cleanup(struct net_device *netdev)
2085 {
2086 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2087 
2088 	/* ensure that transmissions are stopped if called by do_reset */
2089 
2090 	adapter->tx_queues_active = false;
2091 
2092 	/* Ensure complete_tx() and ibmvnic_xmit() see ->tx_queues_active
2093 	 * update so they don't restart a queue after we stop it below.
2094 	 */
2095 	synchronize_rcu();
2096 
2097 	if (test_bit(0, &adapter->resetting))
2098 		netif_tx_disable(netdev);
2099 	else
2100 		netif_tx_stop_all_queues(netdev);
2101 
2102 	ibmvnic_napi_disable(adapter);
2103 	ibmvnic_disable_irqs(adapter);
2104 }
2105 
2106 static int __ibmvnic_close(struct net_device *netdev)
2107 {
2108 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2109 	int rc = 0;
2110 
2111 	adapter->state = VNIC_CLOSING;
2112 	rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2113 	adapter->state = VNIC_CLOSED;
2114 	return rc;
2115 }
2116 
2117 static int ibmvnic_close(struct net_device *netdev)
2118 {
2119 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2120 	int rc;
2121 
2122 	netdev_dbg(netdev, "[S:%s FOP:%d FRR:%d] Closing\n",
2123 		   adapter_state_to_string(adapter->state),
2124 		   adapter->failover_pending,
2125 		   adapter->force_reset_recovery);
2126 
2127 	/* If device failover is pending, just set device state and return.
2128 	 * Device operation will be handled by reset routine.
2129 	 */
2130 	if (adapter->failover_pending) {
2131 		adapter->state = VNIC_CLOSED;
2132 		return 0;
2133 	}
2134 
2135 	rc = __ibmvnic_close(netdev);
2136 	ibmvnic_cleanup(netdev);
2137 	clean_rx_pools(adapter);
2138 	clean_tx_pools(adapter);
2139 
2140 	return rc;
2141 }
2142 
2143 /**
2144  * get_hdr_lens - fills list of L2/L3/L4 hdr lens
2145  * @hdr_field: bitfield determining needed headers
2146  * @skb: socket buffer
2147  * @hdr_len: array of header lengths to be filled
2148  *
2149  * Reads hdr_field to determine which headers are needed by firmware.
2150  * Builds a buffer containing these headers.  Saves individual header
2151  * lengths and total buffer length to be used to build descriptors.
2152  *
2153  * Return: total len of all headers
2154  */
2155 static int get_hdr_lens(u8 hdr_field, struct sk_buff *skb,
2156 			int *hdr_len)
2157 {
2158 	int len = 0;
2159 
2160 
2161 	if ((hdr_field >> 6) & 1) {
2162 		hdr_len[0] = skb_mac_header_len(skb);
2163 		len += hdr_len[0];
2164 	}
2165 
2166 	if ((hdr_field >> 5) & 1) {
2167 		hdr_len[1] = skb_network_header_len(skb);
2168 		len += hdr_len[1];
2169 	}
2170 
2171 	if (!((hdr_field >> 4) & 1))
2172 		return len;
2173 
2174 	if (skb->protocol == htons(ETH_P_IP)) {
2175 		if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2176 			hdr_len[2] = tcp_hdrlen(skb);
2177 		else if (ip_hdr(skb)->protocol == IPPROTO_UDP)
2178 			hdr_len[2] = sizeof(struct udphdr);
2179 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
2180 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2181 			hdr_len[2] = tcp_hdrlen(skb);
2182 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
2183 			hdr_len[2] = sizeof(struct udphdr);
2184 	}
2185 
2186 	return len + hdr_len[2];
2187 }
2188 
2189 /**
2190  * create_hdr_descs - create header and header extension descriptors
2191  * @hdr_field: bitfield determining needed headers
2192  * @hdr_data: buffer containing header data
2193  * @len: length of data buffer
2194  * @hdr_len: array of individual header lengths
2195  * @scrq_arr: descriptor array
2196  *
2197  * Creates header and, if needed, header extension descriptors and
2198  * places them in a descriptor array, scrq_arr
2199  *
2200  * Return: Number of header descs
2201  */
2202 
2203 static int create_hdr_descs(u8 hdr_field, u8 *hdr_data, int len, int *hdr_len,
2204 			    union sub_crq *scrq_arr)
2205 {
2206 	union sub_crq *hdr_desc;
2207 	int tmp_len = len;
2208 	int num_descs = 0;
2209 	u8 *data, *cur;
2210 	int tmp;
2211 
2212 	while (tmp_len > 0) {
2213 		cur = hdr_data + len - tmp_len;
2214 
2215 		hdr_desc = &scrq_arr[num_descs];
2216 		if (num_descs) {
2217 			data = hdr_desc->hdr_ext.data;
2218 			tmp = tmp_len > 29 ? 29 : tmp_len;
2219 			hdr_desc->hdr_ext.first = IBMVNIC_CRQ_CMD;
2220 			hdr_desc->hdr_ext.type = IBMVNIC_HDR_EXT_DESC;
2221 			hdr_desc->hdr_ext.len = tmp;
2222 		} else {
2223 			data = hdr_desc->hdr.data;
2224 			tmp = tmp_len > 24 ? 24 : tmp_len;
2225 			hdr_desc->hdr.first = IBMVNIC_CRQ_CMD;
2226 			hdr_desc->hdr.type = IBMVNIC_HDR_DESC;
2227 			hdr_desc->hdr.len = tmp;
2228 			hdr_desc->hdr.l2_len = (u8)hdr_len[0];
2229 			hdr_desc->hdr.l3_len = cpu_to_be16((u16)hdr_len[1]);
2230 			hdr_desc->hdr.l4_len = (u8)hdr_len[2];
2231 			hdr_desc->hdr.flag = hdr_field << 1;
2232 		}
2233 		memcpy(data, cur, tmp);
2234 		tmp_len -= tmp;
2235 		num_descs++;
2236 	}
2237 
2238 	return num_descs;
2239 }
2240 
2241 /**
2242  * build_hdr_descs_arr - build a header descriptor array
2243  * @skb: tx socket buffer
2244  * @indir_arr: indirect array
2245  * @num_entries: number of descriptors to be sent
2246  * @hdr_field: bit field determining which headers will be sent
2247  *
2248  * This function will build a TX descriptor array with applicable
2249  * L2/L3/L4 packet header descriptors to be sent by send_subcrq_indirect.
2250  */
2251 
2252 static void build_hdr_descs_arr(struct sk_buff *skb,
2253 				union sub_crq *indir_arr,
2254 				int *num_entries, u8 hdr_field)
2255 {
2256 	int hdr_len[3] = {0, 0, 0};
2257 	int tot_len;
2258 
2259 	tot_len = get_hdr_lens(hdr_field, skb, hdr_len);
2260 	*num_entries += create_hdr_descs(hdr_field, skb_mac_header(skb),
2261 					 tot_len, hdr_len, indir_arr + 1);
2262 }
2263 
2264 static int ibmvnic_xmit_workarounds(struct sk_buff *skb,
2265 				    struct net_device *netdev)
2266 {
2267 	/* For some backing devices, mishandling of small packets
2268 	 * can result in a loss of connection or TX stall. Device
2269 	 * architects recommend that no packet should be smaller
2270 	 * than the minimum MTU value provided to the driver, so
2271 	 * pad any packets to that length
2272 	 */
2273 	if (skb->len < netdev->min_mtu)
2274 		return skb_put_padto(skb, netdev->min_mtu);
2275 
2276 	return 0;
2277 }
2278 
2279 static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
2280 					 struct ibmvnic_sub_crq_queue *tx_scrq)
2281 {
2282 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2283 	struct ibmvnic_tx_buff *tx_buff;
2284 	struct ibmvnic_tx_pool *tx_pool;
2285 	union sub_crq tx_scrq_entry;
2286 	int queue_num;
2287 	int entries;
2288 	int index;
2289 	int i;
2290 
2291 	ind_bufp = &tx_scrq->ind_buf;
2292 	entries = (u64)ind_bufp->index;
2293 	queue_num = tx_scrq->pool_index;
2294 
2295 	for (i = entries - 1; i >= 0; --i) {
2296 		tx_scrq_entry = ind_bufp->indir_arr[i];
2297 		if (tx_scrq_entry.v1.type != IBMVNIC_TX_DESC)
2298 			continue;
2299 		index = be32_to_cpu(tx_scrq_entry.v1.correlator);
2300 		if (index & IBMVNIC_TSO_POOL_MASK) {
2301 			tx_pool = &adapter->tso_pool[queue_num];
2302 			index &= ~IBMVNIC_TSO_POOL_MASK;
2303 		} else {
2304 			tx_pool = &adapter->tx_pool[queue_num];
2305 		}
2306 		tx_pool->free_map[tx_pool->consumer_index] = index;
2307 		tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2308 					  tx_pool->num_buffers - 1 :
2309 					  tx_pool->consumer_index - 1;
2310 		tx_buff = &tx_pool->tx_buff[index];
2311 		adapter->netdev->stats.tx_packets--;
2312 		adapter->netdev->stats.tx_bytes -= tx_buff->skb->len;
2313 		adapter->tx_stats_buffers[queue_num].batched_packets--;
2314 		adapter->tx_stats_buffers[queue_num].bytes -=
2315 						tx_buff->skb->len;
2316 		dev_kfree_skb_any(tx_buff->skb);
2317 		tx_buff->skb = NULL;
2318 		adapter->netdev->stats.tx_dropped++;
2319 	}
2320 
2321 	ind_bufp->index = 0;
2322 
2323 	if (atomic_sub_return(entries, &tx_scrq->used) <=
2324 	    (adapter->req_tx_entries_per_subcrq / 2) &&
2325 	    __netif_subqueue_stopped(adapter->netdev, queue_num)) {
2326 		rcu_read_lock();
2327 
2328 		if (adapter->tx_queues_active) {
2329 			netif_wake_subqueue(adapter->netdev, queue_num);
2330 			netdev_dbg(adapter->netdev, "Started queue %d\n",
2331 				   queue_num);
2332 		}
2333 
2334 		rcu_read_unlock();
2335 	}
2336 }
2337 
2338 static int send_subcrq_direct(struct ibmvnic_adapter *adapter,
2339 			      u64 remote_handle, u64 *entry)
2340 {
2341 	unsigned int ua = adapter->vdev->unit_address;
2342 	struct device *dev = &adapter->vdev->dev;
2343 	int rc;
2344 
2345 	/* Make sure the hypervisor sees the complete request */
2346 	dma_wmb();
2347 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ, ua,
2348 				cpu_to_be64(remote_handle),
2349 				cpu_to_be64(entry[0]), cpu_to_be64(entry[1]),
2350 				cpu_to_be64(entry[2]), cpu_to_be64(entry[3]));
2351 
2352 	if (rc)
2353 		print_subcrq_error(dev, rc, __func__);
2354 
2355 	return rc;
2356 }
2357 
2358 static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
2359 				 struct ibmvnic_sub_crq_queue *tx_scrq,
2360 				 bool indirect)
2361 {
2362 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2363 	u64 dma_addr;
2364 	u64 entries;
2365 	u64 handle;
2366 	int rc;
2367 
2368 	ind_bufp = &tx_scrq->ind_buf;
2369 	dma_addr = (u64)ind_bufp->indir_dma;
2370 	entries = (u64)ind_bufp->index;
2371 	handle = tx_scrq->handle;
2372 
2373 	if (!entries)
2374 		return 0;
2375 
2376 	if (indirect)
2377 		rc = send_subcrq_indirect(adapter, handle, dma_addr, entries);
2378 	else
2379 		rc = send_subcrq_direct(adapter, handle,
2380 					(u64 *)ind_bufp->indir_arr);
2381 
2382 	if (rc)
2383 		ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
2384 	else
2385 		ind_bufp->index = 0;
2386 	return rc;
2387 }
2388 
2389 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
2390 {
2391 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2392 	int queue_num = skb_get_queue_mapping(skb);
2393 	u8 *hdrs = (u8 *)&adapter->tx_rx_desc_req;
2394 	struct device *dev = &adapter->vdev->dev;
2395 	struct ibmvnic_ind_xmit_queue *ind_bufp;
2396 	struct ibmvnic_tx_buff *tx_buff = NULL;
2397 	struct ibmvnic_sub_crq_queue *tx_scrq;
2398 	struct ibmvnic_long_term_buff *ltb;
2399 	struct ibmvnic_tx_pool *tx_pool;
2400 	unsigned int tx_send_failed = 0;
2401 	netdev_tx_t ret = NETDEV_TX_OK;
2402 	unsigned int tx_map_failed = 0;
2403 	union sub_crq indir_arr[16];
2404 	unsigned int tx_dropped = 0;
2405 	unsigned int tx_dpackets = 0;
2406 	unsigned int tx_bpackets = 0;
2407 	unsigned int tx_bytes = 0;
2408 	dma_addr_t data_dma_addr;
2409 	struct netdev_queue *txq;
2410 	unsigned long lpar_rc;
2411 	union sub_crq tx_crq;
2412 	unsigned int offset;
2413 	bool use_scrq_send_direct = false;
2414 	int num_entries = 1;
2415 	unsigned char *dst;
2416 	int bufidx = 0;
2417 	u8 proto = 0;
2418 
2419 	/* If a reset is in progress, drop the packet since
2420 	 * the scrqs may get torn down. Otherwise use the
2421 	 * rcu to ensure reset waits for us to complete.
2422 	 */
2423 	rcu_read_lock();
2424 	if (!adapter->tx_queues_active) {
2425 		dev_kfree_skb_any(skb);
2426 
2427 		tx_send_failed++;
2428 		tx_dropped++;
2429 		ret = NETDEV_TX_OK;
2430 		goto out;
2431 	}
2432 
2433 	tx_scrq = adapter->tx_scrq[queue_num];
2434 	txq = netdev_get_tx_queue(netdev, queue_num);
2435 	ind_bufp = &tx_scrq->ind_buf;
2436 
2437 	if (ibmvnic_xmit_workarounds(skb, netdev)) {
2438 		tx_dropped++;
2439 		tx_send_failed++;
2440 		ret = NETDEV_TX_OK;
2441 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2442 		if (lpar_rc != H_SUCCESS)
2443 			goto tx_err;
2444 		goto out;
2445 	}
2446 
2447 	if (skb_is_gso(skb))
2448 		tx_pool = &adapter->tso_pool[queue_num];
2449 	else
2450 		tx_pool = &adapter->tx_pool[queue_num];
2451 
2452 	bufidx = tx_pool->free_map[tx_pool->consumer_index];
2453 
2454 	if (bufidx == IBMVNIC_INVALID_MAP) {
2455 		dev_kfree_skb_any(skb);
2456 		tx_send_failed++;
2457 		tx_dropped++;
2458 		ret = NETDEV_TX_OK;
2459 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2460 		if (lpar_rc != H_SUCCESS)
2461 			goto tx_err;
2462 		goto out;
2463 	}
2464 
2465 	tx_pool->free_map[tx_pool->consumer_index] = IBMVNIC_INVALID_MAP;
2466 
2467 	map_txpool_buf_to_ltb(tx_pool, bufidx, &ltb, &offset);
2468 
2469 	dst = ltb->buff + offset;
2470 	memset(dst, 0, tx_pool->buf_size);
2471 	data_dma_addr = ltb->addr + offset;
2472 
2473 	/* if we are going to send_subcrq_direct this then we need to
2474 	 * update the checksum before copying the data into ltb. Essentially
2475 	 * these packets force disable CSO so that we can guarantee that
2476 	 * FW does not need header info and we can send direct. Also, vnic
2477 	 * server must be able to xmit standard packets without header data
2478 	 */
2479 	if (*hdrs == 0 && !skb_is_gso(skb) &&
2480 	    !ind_bufp->index && !netdev_xmit_more()) {
2481 		use_scrq_send_direct = true;
2482 		if (skb->ip_summed == CHECKSUM_PARTIAL &&
2483 		    skb_checksum_help(skb))
2484 			use_scrq_send_direct = false;
2485 	}
2486 
2487 	if (skb_shinfo(skb)->nr_frags) {
2488 		int cur, i;
2489 
2490 		/* Copy the head */
2491 		skb_copy_from_linear_data(skb, dst, skb_headlen(skb));
2492 		cur = skb_headlen(skb);
2493 
2494 		/* Copy the frags */
2495 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2496 			const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2497 
2498 			memcpy(dst + cur, skb_frag_address(frag),
2499 			       skb_frag_size(frag));
2500 			cur += skb_frag_size(frag);
2501 		}
2502 	} else {
2503 		skb_copy_from_linear_data(skb, dst, skb->len);
2504 	}
2505 
2506 	tx_pool->consumer_index =
2507 	    (tx_pool->consumer_index + 1) % tx_pool->num_buffers;
2508 
2509 	tx_buff = &tx_pool->tx_buff[bufidx];
2510 
2511 	/* Sanity checks on our free map to make sure it points to an index
2512 	 * that is not being occupied by another skb. If skb memory is
2513 	 * not freed then we see congestion control kick in and halt tx.
2514 	 */
2515 	if (unlikely(tx_buff->skb)) {
2516 		dev_warn_ratelimited(dev, "TX free map points to untracked skb (%s %d idx=%d)\n",
2517 				     skb_is_gso(skb) ? "tso_pool" : "tx_pool",
2518 				     queue_num, bufidx);
2519 		dev_kfree_skb_any(tx_buff->skb);
2520 	}
2521 
2522 	tx_buff->skb = skb;
2523 	tx_buff->index = bufidx;
2524 	tx_buff->pool_index = queue_num;
2525 
2526 	memset(&tx_crq, 0, sizeof(tx_crq));
2527 	tx_crq.v1.first = IBMVNIC_CRQ_CMD;
2528 	tx_crq.v1.type = IBMVNIC_TX_DESC;
2529 	tx_crq.v1.n_crq_elem = 1;
2530 	tx_crq.v1.n_sge = 1;
2531 	tx_crq.v1.flags1 = IBMVNIC_TX_COMP_NEEDED;
2532 
2533 	if (skb_is_gso(skb))
2534 		tx_crq.v1.correlator =
2535 			cpu_to_be32(bufidx | IBMVNIC_TSO_POOL_MASK);
2536 	else
2537 		tx_crq.v1.correlator = cpu_to_be32(bufidx);
2538 	tx_crq.v1.dma_reg = cpu_to_be16(ltb->map_id);
2539 	tx_crq.v1.sge_len = cpu_to_be32(skb->len);
2540 	tx_crq.v1.ioba = cpu_to_be64(data_dma_addr);
2541 
2542 	if (adapter->vlan_header_insertion && skb_vlan_tag_present(skb)) {
2543 		tx_crq.v1.flags2 |= IBMVNIC_TX_VLAN_INSERT;
2544 		tx_crq.v1.vlan_id = cpu_to_be16(skb->vlan_tci);
2545 	}
2546 
2547 	if (skb->protocol == htons(ETH_P_IP)) {
2548 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV4;
2549 		proto = ip_hdr(skb)->protocol;
2550 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
2551 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_IPV6;
2552 		proto = ipv6_hdr(skb)->nexthdr;
2553 	}
2554 
2555 	if (proto == IPPROTO_TCP)
2556 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_TCP;
2557 	else if (proto == IPPROTO_UDP)
2558 		tx_crq.v1.flags1 |= IBMVNIC_TX_PROT_UDP;
2559 
2560 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
2561 		tx_crq.v1.flags1 |= IBMVNIC_TX_CHKSUM_OFFLOAD;
2562 		hdrs += 2;
2563 	}
2564 	if (skb_is_gso(skb)) {
2565 		tx_crq.v1.flags1 |= IBMVNIC_TX_LSO;
2566 		tx_crq.v1.mss = cpu_to_be16(skb_shinfo(skb)->gso_size);
2567 		hdrs += 2;
2568 	} else if (use_scrq_send_direct) {
2569 		/* See above comment, CSO disabled with direct xmit */
2570 		tx_crq.v1.flags1 &= ~(IBMVNIC_TX_CHKSUM_OFFLOAD);
2571 		ind_bufp->index = 1;
2572 		tx_buff->num_entries = 1;
2573 		netdev_tx_sent_queue(txq, skb->len);
2574 		ind_bufp->indir_arr[0] = tx_crq;
2575 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, false);
2576 		if (lpar_rc != H_SUCCESS)
2577 			goto tx_err;
2578 
2579 		tx_dpackets++;
2580 		goto early_exit;
2581 	}
2582 
2583 	if ((*hdrs >> 7) & 1)
2584 		build_hdr_descs_arr(skb, indir_arr, &num_entries, *hdrs);
2585 
2586 	tx_crq.v1.n_crq_elem = num_entries;
2587 	tx_buff->num_entries = num_entries;
2588 	/* flush buffer if current entry can not fit */
2589 	if (num_entries + ind_bufp->index > IBMVNIC_MAX_IND_DESCS) {
2590 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2591 		if (lpar_rc != H_SUCCESS)
2592 			goto tx_flush_err;
2593 	}
2594 
2595 	indir_arr[0] = tx_crq;
2596 	memcpy(&ind_bufp->indir_arr[ind_bufp->index], &indir_arr[0],
2597 	       num_entries * sizeof(struct ibmvnic_generic_scrq));
2598 
2599 	ind_bufp->index += num_entries;
2600 	if (__netdev_tx_sent_queue(txq, skb->len,
2601 				   netdev_xmit_more() &&
2602 				   ind_bufp->index < IBMVNIC_MAX_IND_DESCS)) {
2603 		lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq, true);
2604 		if (lpar_rc != H_SUCCESS)
2605 			goto tx_err;
2606 	}
2607 
2608 	tx_bpackets++;
2609 
2610 early_exit:
2611 	if (atomic_add_return(num_entries, &tx_scrq->used)
2612 					>= adapter->req_tx_entries_per_subcrq) {
2613 		netdev_dbg(netdev, "Stopping queue %d\n", queue_num);
2614 		netif_stop_subqueue(netdev, queue_num);
2615 	}
2616 
2617 	tx_bytes += skb->len;
2618 	txq_trans_cond_update(txq);
2619 	ret = NETDEV_TX_OK;
2620 	goto out;
2621 
2622 tx_flush_err:
2623 	dev_kfree_skb_any(skb);
2624 	tx_buff->skb = NULL;
2625 	tx_pool->consumer_index = tx_pool->consumer_index == 0 ?
2626 				  tx_pool->num_buffers - 1 :
2627 				  tx_pool->consumer_index - 1;
2628 	tx_dropped++;
2629 tx_err:
2630 	if (lpar_rc != H_CLOSED && lpar_rc != H_PARAMETER)
2631 		dev_err_ratelimited(dev, "tx: send failed\n");
2632 
2633 	if (lpar_rc == H_CLOSED || adapter->failover_pending) {
2634 		/* Disable TX and report carrier off if queue is closed
2635 		 * or pending failover.
2636 		 * Firmware guarantees that a signal will be sent to the
2637 		 * driver, triggering a reset or some other action.
2638 		 */
2639 		netif_tx_stop_all_queues(netdev);
2640 		netif_carrier_off(netdev);
2641 	}
2642 out:
2643 	rcu_read_unlock();
2644 	netdev->stats.tx_dropped += tx_dropped;
2645 	netdev->stats.tx_bytes += tx_bytes;
2646 	netdev->stats.tx_packets += tx_bpackets + tx_dpackets;
2647 	adapter->tx_send_failed += tx_send_failed;
2648 	adapter->tx_map_failed += tx_map_failed;
2649 	adapter->tx_stats_buffers[queue_num].batched_packets += tx_bpackets;
2650 	adapter->tx_stats_buffers[queue_num].direct_packets += tx_dpackets;
2651 	adapter->tx_stats_buffers[queue_num].bytes += tx_bytes;
2652 	adapter->tx_stats_buffers[queue_num].dropped_packets += tx_dropped;
2653 
2654 	return ret;
2655 }
2656 
2657 static void ibmvnic_set_multi(struct net_device *netdev)
2658 {
2659 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2660 	struct netdev_hw_addr *ha;
2661 	union ibmvnic_crq crq;
2662 
2663 	memset(&crq, 0, sizeof(crq));
2664 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
2665 	crq.request_capability.cmd = REQUEST_CAPABILITY;
2666 
2667 	if (netdev->flags & IFF_PROMISC) {
2668 		if (!adapter->promisc_supported)
2669 			return;
2670 	} else {
2671 		if (netdev->flags & IFF_ALLMULTI) {
2672 			/* Accept all multicast */
2673 			memset(&crq, 0, sizeof(crq));
2674 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2675 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2676 			crq.multicast_ctrl.flags = IBMVNIC_ENABLE_ALL;
2677 			ibmvnic_send_crq(adapter, &crq);
2678 		} else if (netdev_mc_empty(netdev)) {
2679 			/* Reject all multicast */
2680 			memset(&crq, 0, sizeof(crq));
2681 			crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2682 			crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2683 			crq.multicast_ctrl.flags = IBMVNIC_DISABLE_ALL;
2684 			ibmvnic_send_crq(adapter, &crq);
2685 		} else {
2686 			/* Accept one or more multicast(s) */
2687 			netdev_for_each_mc_addr(ha, netdev) {
2688 				memset(&crq, 0, sizeof(crq));
2689 				crq.multicast_ctrl.first = IBMVNIC_CRQ_CMD;
2690 				crq.multicast_ctrl.cmd = MULTICAST_CTRL;
2691 				crq.multicast_ctrl.flags = IBMVNIC_ENABLE_MC;
2692 				ether_addr_copy(&crq.multicast_ctrl.mac_addr[0],
2693 						ha->addr);
2694 				ibmvnic_send_crq(adapter, &crq);
2695 			}
2696 		}
2697 	}
2698 }
2699 
2700 static int __ibmvnic_set_mac(struct net_device *netdev, u8 *dev_addr)
2701 {
2702 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2703 	union ibmvnic_crq crq;
2704 	int rc;
2705 
2706 	if (!is_valid_ether_addr(dev_addr)) {
2707 		rc = -EADDRNOTAVAIL;
2708 		goto err;
2709 	}
2710 
2711 	memset(&crq, 0, sizeof(crq));
2712 	crq.change_mac_addr.first = IBMVNIC_CRQ_CMD;
2713 	crq.change_mac_addr.cmd = CHANGE_MAC_ADDR;
2714 	ether_addr_copy(&crq.change_mac_addr.mac_addr[0], dev_addr);
2715 
2716 	mutex_lock(&adapter->fw_lock);
2717 	adapter->fw_done_rc = 0;
2718 	reinit_completion(&adapter->fw_done);
2719 
2720 	rc = ibmvnic_send_crq(adapter, &crq);
2721 	if (rc) {
2722 		rc = -EIO;
2723 		mutex_unlock(&adapter->fw_lock);
2724 		goto err;
2725 	}
2726 
2727 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
2728 	/* netdev->dev_addr is changed in handle_change_mac_rsp function */
2729 	if (rc || adapter->fw_done_rc) {
2730 		rc = -EIO;
2731 		mutex_unlock(&adapter->fw_lock);
2732 		goto err;
2733 	}
2734 	mutex_unlock(&adapter->fw_lock);
2735 	return 0;
2736 err:
2737 	ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2738 	return rc;
2739 }
2740 
2741 static int ibmvnic_set_mac(struct net_device *netdev, void *p)
2742 {
2743 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
2744 	struct sockaddr *addr = p;
2745 	int rc;
2746 
2747 	rc = 0;
2748 	if (!is_valid_ether_addr(addr->sa_data))
2749 		return -EADDRNOTAVAIL;
2750 
2751 	ether_addr_copy(adapter->mac_addr, addr->sa_data);
2752 	if (adapter->state != VNIC_PROBED)
2753 		rc = __ibmvnic_set_mac(netdev, addr->sa_data);
2754 
2755 	return rc;
2756 }
2757 
2758 static const char *reset_reason_to_string(enum ibmvnic_reset_reason reason)
2759 {
2760 	switch (reason) {
2761 	case VNIC_RESET_FAILOVER:
2762 		return "FAILOVER";
2763 	case VNIC_RESET_MOBILITY:
2764 		return "MOBILITY";
2765 	case VNIC_RESET_FATAL:
2766 		return "FATAL";
2767 	case VNIC_RESET_NON_FATAL:
2768 		return "NON_FATAL";
2769 	case VNIC_RESET_TIMEOUT:
2770 		return "TIMEOUT";
2771 	case VNIC_RESET_CHANGE_PARAM:
2772 		return "CHANGE_PARAM";
2773 	case VNIC_RESET_PASSIVE_INIT:
2774 		return "PASSIVE_INIT";
2775 	}
2776 	return "UNKNOWN";
2777 }
2778 
2779 /*
2780  * Initialize the init_done completion and return code values. We
2781  * can get a transport event just after registering the CRQ and the
2782  * tasklet will use this to communicate the transport event. To ensure
2783  * we don't miss the notification/error, initialize these _before_
2784  * regisering the CRQ.
2785  */
2786 static inline void reinit_init_done(struct ibmvnic_adapter *adapter)
2787 {
2788 	reinit_completion(&adapter->init_done);
2789 	adapter->init_done_rc = 0;
2790 }
2791 
2792 /*
2793  * do_reset returns zero if we are able to keep processing reset events, or
2794  * non-zero if we hit a fatal error and must halt.
2795  */
2796 static int do_reset(struct ibmvnic_adapter *adapter,
2797 		    struct ibmvnic_rwi *rwi, u32 reset_state)
2798 {
2799 	struct net_device *netdev = adapter->netdev;
2800 	u64 old_num_rx_queues, old_num_tx_queues;
2801 	u64 old_num_rx_slots, old_num_tx_slots;
2802 	int rc;
2803 
2804 	netdev_dbg(adapter->netdev,
2805 		   "[S:%s FOP:%d] Reset reason: %s, reset_state: %s\n",
2806 		   adapter_state_to_string(adapter->state),
2807 		   adapter->failover_pending,
2808 		   reset_reason_to_string(rwi->reset_reason),
2809 		   adapter_state_to_string(reset_state));
2810 
2811 	adapter->reset_reason = rwi->reset_reason;
2812 	/* requestor of VNIC_RESET_CHANGE_PARAM already has the rtnl lock */
2813 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
2814 		rtnl_lock();
2815 
2816 	/* Now that we have the rtnl lock, clear any pending failover.
2817 	 * This will ensure ibmvnic_open() has either completed or will
2818 	 * block until failover is complete.
2819 	 */
2820 	if (rwi->reset_reason == VNIC_RESET_FAILOVER)
2821 		adapter->failover_pending = false;
2822 
2823 	/* read the state and check (again) after getting rtnl */
2824 	reset_state = adapter->state;
2825 
2826 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
2827 		rc = -EBUSY;
2828 		goto out;
2829 	}
2830 
2831 	netif_carrier_off(netdev);
2832 
2833 	old_num_rx_queues = adapter->req_rx_queues;
2834 	old_num_tx_queues = adapter->req_tx_queues;
2835 	old_num_rx_slots = adapter->req_rx_add_entries_per_subcrq;
2836 	old_num_tx_slots = adapter->req_tx_entries_per_subcrq;
2837 
2838 	ibmvnic_cleanup(netdev);
2839 
2840 	if (reset_state == VNIC_OPEN &&
2841 	    adapter->reset_reason != VNIC_RESET_MOBILITY &&
2842 	    adapter->reset_reason != VNIC_RESET_FAILOVER) {
2843 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2844 			rc = __ibmvnic_close(netdev);
2845 			if (rc)
2846 				goto out;
2847 		} else {
2848 			adapter->state = VNIC_CLOSING;
2849 
2850 			/* Release the RTNL lock before link state change and
2851 			 * re-acquire after the link state change to allow
2852 			 * linkwatch_event to grab the RTNL lock and run during
2853 			 * a reset.
2854 			 */
2855 			rtnl_unlock();
2856 			rc = set_link_state(adapter, IBMVNIC_LOGICAL_LNK_DN);
2857 			rtnl_lock();
2858 			if (rc)
2859 				goto out;
2860 
2861 			if (adapter->state == VNIC_OPEN) {
2862 				/* When we dropped rtnl, ibmvnic_open() got
2863 				 * it and noticed that we are resetting and
2864 				 * set the adapter state to OPEN. Update our
2865 				 * new "target" state, and resume the reset
2866 				 * from VNIC_CLOSING state.
2867 				 */
2868 				netdev_dbg(netdev,
2869 					   "Open changed state from %s, updating.\n",
2870 					   adapter_state_to_string(reset_state));
2871 				reset_state = VNIC_OPEN;
2872 				adapter->state = VNIC_CLOSING;
2873 			}
2874 
2875 			if (adapter->state != VNIC_CLOSING) {
2876 				/* If someone else changed the adapter state
2877 				 * when we dropped the rtnl, fail the reset
2878 				 */
2879 				rc = -EAGAIN;
2880 				goto out;
2881 			}
2882 			adapter->state = VNIC_CLOSED;
2883 		}
2884 	}
2885 
2886 	if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2887 		release_resources(adapter);
2888 		release_sub_crqs(adapter, 1);
2889 		release_crq_queue(adapter);
2890 	}
2891 
2892 	if (adapter->reset_reason != VNIC_RESET_NON_FATAL) {
2893 		/* remove the closed state so when we call open it appears
2894 		 * we are coming from the probed state.
2895 		 */
2896 		adapter->state = VNIC_PROBED;
2897 
2898 		reinit_init_done(adapter);
2899 
2900 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2901 			rc = init_crq_queue(adapter);
2902 		} else if (adapter->reset_reason == VNIC_RESET_MOBILITY) {
2903 			rc = ibmvnic_reenable_crq_queue(adapter);
2904 			release_sub_crqs(adapter, 1);
2905 		} else {
2906 			rc = ibmvnic_reset_crq(adapter);
2907 			if (rc == H_CLOSED || rc == H_SUCCESS) {
2908 				rc = vio_enable_interrupts(adapter->vdev);
2909 				if (rc)
2910 					netdev_err(adapter->netdev,
2911 						   "Reset failed to enable interrupts. rc=%d\n",
2912 						   rc);
2913 			}
2914 		}
2915 
2916 		if (rc) {
2917 			netdev_err(adapter->netdev,
2918 				   "Reset couldn't initialize crq. rc=%d\n", rc);
2919 			goto out;
2920 		}
2921 
2922 		rc = ibmvnic_reset_init(adapter, true);
2923 		if (rc)
2924 			goto out;
2925 
2926 		/* If the adapter was in PROBE or DOWN state prior to the reset,
2927 		 * exit here.
2928 		 */
2929 		if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN) {
2930 			rc = 0;
2931 			goto out;
2932 		}
2933 
2934 		rc = ibmvnic_login(netdev);
2935 		if (rc)
2936 			goto out;
2937 
2938 		if (adapter->reset_reason == VNIC_RESET_CHANGE_PARAM) {
2939 			rc = init_resources(adapter);
2940 			if (rc)
2941 				goto out;
2942 		} else if (adapter->req_rx_queues != old_num_rx_queues ||
2943 		    adapter->req_tx_queues != old_num_tx_queues ||
2944 		    adapter->req_rx_add_entries_per_subcrq !=
2945 		    old_num_rx_slots ||
2946 		    adapter->req_tx_entries_per_subcrq !=
2947 		    old_num_tx_slots ||
2948 		    !adapter->rx_pool ||
2949 		    !adapter->tso_pool ||
2950 		    !adapter->tx_pool) {
2951 			release_napi(adapter);
2952 			release_vpd_data(adapter);
2953 
2954 			rc = init_resources(adapter);
2955 			if (rc)
2956 				goto out;
2957 
2958 		} else {
2959 			rc = init_tx_pools(netdev);
2960 			if (rc) {
2961 				netdev_dbg(netdev,
2962 					   "init tx pools failed (%d)\n",
2963 					   rc);
2964 				goto out;
2965 			}
2966 
2967 			rc = init_rx_pools(netdev);
2968 			if (rc) {
2969 				netdev_dbg(netdev,
2970 					   "init rx pools failed (%d)\n",
2971 					   rc);
2972 				goto out;
2973 			}
2974 		}
2975 		ibmvnic_disable_irqs(adapter);
2976 	}
2977 	adapter->state = VNIC_CLOSED;
2978 
2979 	if (reset_state == VNIC_CLOSED) {
2980 		rc = 0;
2981 		goto out;
2982 	}
2983 
2984 	rc = __ibmvnic_open(netdev);
2985 	if (rc) {
2986 		rc = IBMVNIC_OPEN_FAILED;
2987 		goto out;
2988 	}
2989 
2990 	/* refresh device's multicast list */
2991 	ibmvnic_set_multi(netdev);
2992 
2993 	if (adapter->reset_reason == VNIC_RESET_FAILOVER ||
2994 	    adapter->reset_reason == VNIC_RESET_MOBILITY)
2995 		__netdev_notify_peers(netdev);
2996 
2997 	rc = 0;
2998 
2999 out:
3000 	/* restore the adapter state if reset failed */
3001 	if (rc)
3002 		adapter->state = reset_state;
3003 	/* requestor of VNIC_RESET_CHANGE_PARAM should still hold the rtnl lock */
3004 	if (!(adapter->reset_reason == VNIC_RESET_CHANGE_PARAM))
3005 		rtnl_unlock();
3006 
3007 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Reset done, rc %d\n",
3008 		   adapter_state_to_string(adapter->state),
3009 		   adapter->failover_pending, rc);
3010 	return rc;
3011 }
3012 
3013 static int do_hard_reset(struct ibmvnic_adapter *adapter,
3014 			 struct ibmvnic_rwi *rwi, u32 reset_state)
3015 {
3016 	struct net_device *netdev = adapter->netdev;
3017 	int rc;
3018 
3019 	netdev_dbg(adapter->netdev, "Hard resetting driver (%s)\n",
3020 		   reset_reason_to_string(rwi->reset_reason));
3021 
3022 	/* read the state and check (again) after getting rtnl */
3023 	reset_state = adapter->state;
3024 
3025 	if (reset_state == VNIC_REMOVING || reset_state == VNIC_REMOVED) {
3026 		rc = -EBUSY;
3027 		goto out;
3028 	}
3029 
3030 	netif_carrier_off(netdev);
3031 	adapter->reset_reason = rwi->reset_reason;
3032 
3033 	ibmvnic_cleanup(netdev);
3034 	release_resources(adapter);
3035 	release_sub_crqs(adapter, 0);
3036 	release_crq_queue(adapter);
3037 
3038 	/* remove the closed state so when we call open it appears
3039 	 * we are coming from the probed state.
3040 	 */
3041 	adapter->state = VNIC_PROBED;
3042 
3043 	reinit_init_done(adapter);
3044 
3045 	rc = init_crq_queue(adapter);
3046 	if (rc) {
3047 		netdev_err(adapter->netdev,
3048 			   "Couldn't initialize crq. rc=%d\n", rc);
3049 		goto out;
3050 	}
3051 
3052 	rc = ibmvnic_reset_init(adapter, false);
3053 	if (rc)
3054 		goto out;
3055 
3056 	/* If the adapter was in PROBE or DOWN state prior to the reset,
3057 	 * exit here.
3058 	 */
3059 	if (reset_state == VNIC_PROBED || reset_state == VNIC_DOWN)
3060 		goto out;
3061 
3062 	rc = ibmvnic_login(netdev);
3063 	if (rc)
3064 		goto out;
3065 
3066 	rc = init_resources(adapter);
3067 	if (rc)
3068 		goto out;
3069 
3070 	ibmvnic_disable_irqs(adapter);
3071 	adapter->state = VNIC_CLOSED;
3072 
3073 	if (reset_state == VNIC_CLOSED)
3074 		goto out;
3075 
3076 	rc = __ibmvnic_open(netdev);
3077 	if (rc) {
3078 		rc = IBMVNIC_OPEN_FAILED;
3079 		goto out;
3080 	}
3081 
3082 	__netdev_notify_peers(netdev);
3083 out:
3084 	/* restore adapter state if reset failed */
3085 	if (rc)
3086 		adapter->state = reset_state;
3087 	netdev_dbg(adapter->netdev, "[S:%s FOP:%d] Hard reset done, rc %d\n",
3088 		   adapter_state_to_string(adapter->state),
3089 		   adapter->failover_pending, rc);
3090 	return rc;
3091 }
3092 
3093 static struct ibmvnic_rwi *get_next_rwi(struct ibmvnic_adapter *adapter)
3094 {
3095 	struct ibmvnic_rwi *rwi;
3096 	unsigned long flags;
3097 
3098 	spin_lock_irqsave(&adapter->rwi_lock, flags);
3099 
3100 	if (!list_empty(&adapter->rwi_list)) {
3101 		rwi = list_first_entry(&adapter->rwi_list, struct ibmvnic_rwi,
3102 				       list);
3103 		list_del(&rwi->list);
3104 	} else {
3105 		rwi = NULL;
3106 	}
3107 
3108 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
3109 	return rwi;
3110 }
3111 
3112 /**
3113  * do_passive_init - complete probing when partner device is detected.
3114  * @adapter: ibmvnic_adapter struct
3115  *
3116  * If the ibmvnic device does not have a partner device to communicate with at boot
3117  * and that partner device comes online at a later time, this function is called
3118  * to complete the initialization process of ibmvnic device.
3119  * Caller is expected to hold rtnl_lock().
3120  *
3121  * Returns non-zero if sub-CRQs are not initialized properly leaving the device
3122  * in the down state.
3123  * Returns 0 upon success and the device is in PROBED state.
3124  */
3125 
3126 static int do_passive_init(struct ibmvnic_adapter *adapter)
3127 {
3128 	unsigned long timeout = msecs_to_jiffies(30000);
3129 	struct net_device *netdev = adapter->netdev;
3130 	struct device *dev = &adapter->vdev->dev;
3131 	int rc;
3132 
3133 	netdev_dbg(netdev, "Partner device found, probing.\n");
3134 
3135 	adapter->state = VNIC_PROBING;
3136 	reinit_completion(&adapter->init_done);
3137 	adapter->init_done_rc = 0;
3138 	adapter->crq.active = true;
3139 
3140 	rc = send_crq_init_complete(adapter);
3141 	if (rc)
3142 		goto out;
3143 
3144 	rc = send_version_xchg(adapter);
3145 	if (rc)
3146 		netdev_dbg(adapter->netdev, "send_version_xchg failed, rc=%d\n", rc);
3147 
3148 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
3149 		dev_err(dev, "Initialization sequence timed out\n");
3150 		rc = -ETIMEDOUT;
3151 		goto out;
3152 	}
3153 
3154 	rc = init_sub_crqs(adapter);
3155 	if (rc) {
3156 		dev_err(dev, "Initialization of sub crqs failed, rc=%d\n", rc);
3157 		goto out;
3158 	}
3159 
3160 	rc = init_sub_crq_irqs(adapter);
3161 	if (rc) {
3162 		dev_err(dev, "Failed to initialize sub crq irqs\n, rc=%d", rc);
3163 		goto init_failed;
3164 	}
3165 
3166 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
3167 	netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
3168 	netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
3169 
3170 	adapter->state = VNIC_PROBED;
3171 	netdev_dbg(netdev, "Probed successfully. Waiting for signal from partner device.\n");
3172 
3173 	return 0;
3174 
3175 init_failed:
3176 	release_sub_crqs(adapter, 1);
3177 out:
3178 	adapter->state = VNIC_DOWN;
3179 	return rc;
3180 }
3181 
3182 static void __ibmvnic_reset(struct work_struct *work)
3183 {
3184 	struct ibmvnic_adapter *adapter;
3185 	unsigned int timeout = 5000;
3186 	struct ibmvnic_rwi *tmprwi;
3187 	bool saved_state = false;
3188 	struct ibmvnic_rwi *rwi;
3189 	unsigned long flags;
3190 	struct device *dev;
3191 	bool need_reset;
3192 	int num_fails = 0;
3193 	u32 reset_state;
3194 	int rc = 0;
3195 
3196 	adapter = container_of(work, struct ibmvnic_adapter, ibmvnic_reset);
3197 		dev = &adapter->vdev->dev;
3198 
3199 	/* Wait for ibmvnic_probe() to complete. If probe is taking too long
3200 	 * or if another reset is in progress, defer work for now. If probe
3201 	 * eventually fails it will flush and terminate our work.
3202 	 *
3203 	 * Three possibilities here:
3204 	 * 1. Adpater being removed  - just return
3205 	 * 2. Timed out on probe or another reset in progress - delay the work
3206 	 * 3. Completed probe - perform any resets in queue
3207 	 */
3208 	if (adapter->state == VNIC_PROBING &&
3209 	    !wait_for_completion_timeout(&adapter->probe_done, timeout)) {
3210 		dev_err(dev, "Reset thread timed out on probe");
3211 		queue_delayed_work(system_long_wq,
3212 				   &adapter->ibmvnic_delayed_reset,
3213 				   IBMVNIC_RESET_DELAY);
3214 		return;
3215 	}
3216 
3217 	/* adapter is done with probe (i.e state is never VNIC_PROBING now) */
3218 	if (adapter->state == VNIC_REMOVING)
3219 		return;
3220 
3221 	/* ->rwi_list is stable now (no one else is removing entries) */
3222 
3223 	/* ibmvnic_probe() may have purged the reset queue after we were
3224 	 * scheduled to process a reset so there maybe no resets to process.
3225 	 * Before setting the ->resetting bit though, we have to make sure
3226 	 * that there is infact a reset to process. Otherwise we may race
3227 	 * with ibmvnic_open() and end up leaving the vnic down:
3228 	 *
3229 	 *	__ibmvnic_reset()	    ibmvnic_open()
3230 	 *	-----------------	    --------------
3231 	 *
3232 	 *  set ->resetting bit
3233 	 *  				find ->resetting bit is set
3234 	 *  				set ->state to IBMVNIC_OPEN (i.e
3235 	 *  				assume reset will open device)
3236 	 *  				return
3237 	 *  find reset queue empty
3238 	 *  return
3239 	 *
3240 	 *  	Neither performed vnic login/open and vnic stays down
3241 	 *
3242 	 * If we hold the lock and conditionally set the bit, either we
3243 	 * or ibmvnic_open() will complete the open.
3244 	 */
3245 	need_reset = false;
3246 	spin_lock(&adapter->rwi_lock);
3247 	if (!list_empty(&adapter->rwi_list)) {
3248 		if (test_and_set_bit_lock(0, &adapter->resetting)) {
3249 			queue_delayed_work(system_long_wq,
3250 					   &adapter->ibmvnic_delayed_reset,
3251 					   IBMVNIC_RESET_DELAY);
3252 		} else {
3253 			need_reset = true;
3254 		}
3255 	}
3256 	spin_unlock(&adapter->rwi_lock);
3257 
3258 	if (!need_reset)
3259 		return;
3260 
3261 	rwi = get_next_rwi(adapter);
3262 	while (rwi) {
3263 		spin_lock_irqsave(&adapter->state_lock, flags);
3264 
3265 		if (adapter->state == VNIC_REMOVING ||
3266 		    adapter->state == VNIC_REMOVED) {
3267 			spin_unlock_irqrestore(&adapter->state_lock, flags);
3268 			kfree(rwi);
3269 			rc = EBUSY;
3270 			break;
3271 		}
3272 
3273 		if (!saved_state) {
3274 			reset_state = adapter->state;
3275 			saved_state = true;
3276 		}
3277 		spin_unlock_irqrestore(&adapter->state_lock, flags);
3278 
3279 		if (rwi->reset_reason == VNIC_RESET_PASSIVE_INIT) {
3280 			rtnl_lock();
3281 			rc = do_passive_init(adapter);
3282 			rtnl_unlock();
3283 			if (!rc)
3284 				netif_carrier_on(adapter->netdev);
3285 		} else if (adapter->force_reset_recovery) {
3286 			/* Since we are doing a hard reset now, clear the
3287 			 * failover_pending flag so we don't ignore any
3288 			 * future MOBILITY or other resets.
3289 			 */
3290 			adapter->failover_pending = false;
3291 
3292 			/* Transport event occurred during previous reset */
3293 			if (adapter->wait_for_reset) {
3294 				/* Previous was CHANGE_PARAM; caller locked */
3295 				adapter->force_reset_recovery = false;
3296 				rc = do_hard_reset(adapter, rwi, reset_state);
3297 			} else {
3298 				rtnl_lock();
3299 				adapter->force_reset_recovery = false;
3300 				rc = do_hard_reset(adapter, rwi, reset_state);
3301 				rtnl_unlock();
3302 			}
3303 			if (rc)
3304 				num_fails++;
3305 			else
3306 				num_fails = 0;
3307 
3308 			/* If auto-priority-failover is enabled we can get
3309 			 * back to back failovers during resets, resulting
3310 			 * in at least two failed resets (from high-priority
3311 			 * backing device to low-priority one and then back)
3312 			 * If resets continue to fail beyond that, give the
3313 			 * adapter some time to settle down before retrying.
3314 			 */
3315 			if (num_fails >= 3) {
3316 				netdev_dbg(adapter->netdev,
3317 					   "[S:%s] Hard reset failed %d times, waiting 60 secs\n",
3318 					   adapter_state_to_string(adapter->state),
3319 					   num_fails);
3320 				set_current_state(TASK_UNINTERRUPTIBLE);
3321 				schedule_timeout(60 * HZ);
3322 			}
3323 		} else {
3324 			rc = do_reset(adapter, rwi, reset_state);
3325 		}
3326 		tmprwi = rwi;
3327 		adapter->last_reset_time = jiffies;
3328 
3329 		if (rc)
3330 			netdev_dbg(adapter->netdev, "Reset failed, rc=%d\n", rc);
3331 
3332 		rwi = get_next_rwi(adapter);
3333 
3334 		/*
3335 		 * If there are no resets queued and the previous reset failed,
3336 		 * the adapter would be in an undefined state. So retry the
3337 		 * previous reset as a hard reset.
3338 		 *
3339 		 * Else, free the previous rwi and, if there is another reset
3340 		 * queued, process the new reset even if previous reset failed
3341 		 * (the previous reset could have failed because of a fail
3342 		 * over for instance, so process the fail over).
3343 		 */
3344 		if (!rwi && rc)
3345 			rwi = tmprwi;
3346 		else
3347 			kfree(tmprwi);
3348 
3349 		if (rwi && (rwi->reset_reason == VNIC_RESET_FAILOVER ||
3350 			    rwi->reset_reason == VNIC_RESET_MOBILITY || rc))
3351 			adapter->force_reset_recovery = true;
3352 	}
3353 
3354 	if (adapter->wait_for_reset) {
3355 		adapter->reset_done_rc = rc;
3356 		complete(&adapter->reset_done);
3357 	}
3358 
3359 	clear_bit_unlock(0, &adapter->resetting);
3360 
3361 	netdev_dbg(adapter->netdev,
3362 		   "[S:%s FRR:%d WFR:%d] Done processing resets\n",
3363 		   adapter_state_to_string(adapter->state),
3364 		   adapter->force_reset_recovery,
3365 		   adapter->wait_for_reset);
3366 }
3367 
3368 static void __ibmvnic_delayed_reset(struct work_struct *work)
3369 {
3370 	struct ibmvnic_adapter *adapter;
3371 
3372 	adapter = container_of(work, struct ibmvnic_adapter,
3373 			       ibmvnic_delayed_reset.work);
3374 	__ibmvnic_reset(&adapter->ibmvnic_reset);
3375 }
3376 
3377 static void flush_reset_queue(struct ibmvnic_adapter *adapter)
3378 {
3379 	struct list_head *entry, *tmp_entry;
3380 
3381 	if (!list_empty(&adapter->rwi_list)) {
3382 		list_for_each_safe(entry, tmp_entry, &adapter->rwi_list) {
3383 			list_del(entry);
3384 			kfree(list_entry(entry, struct ibmvnic_rwi, list));
3385 		}
3386 	}
3387 }
3388 
3389 static int ibmvnic_reset(struct ibmvnic_adapter *adapter,
3390 			 enum ibmvnic_reset_reason reason)
3391 {
3392 	struct net_device *netdev = adapter->netdev;
3393 	struct ibmvnic_rwi *rwi, *tmp;
3394 	unsigned long flags;
3395 	int ret;
3396 
3397 	spin_lock_irqsave(&adapter->rwi_lock, flags);
3398 
3399 	/* If failover is pending don't schedule any other reset.
3400 	 * Instead let the failover complete. If there is already a
3401 	 * a failover reset scheduled, we will detect and drop the
3402 	 * duplicate reset when walking the ->rwi_list below.
3403 	 */
3404 	if (adapter->state == VNIC_REMOVING ||
3405 	    adapter->state == VNIC_REMOVED ||
3406 	    (adapter->failover_pending && reason != VNIC_RESET_FAILOVER)) {
3407 		ret = EBUSY;
3408 		netdev_dbg(netdev, "Adapter removing or pending failover, skipping reset\n");
3409 		goto err;
3410 	}
3411 
3412 	list_for_each_entry(tmp, &adapter->rwi_list, list) {
3413 		if (tmp->reset_reason == reason) {
3414 			netdev_dbg(netdev, "Skipping matching reset, reason=%s\n",
3415 				   reset_reason_to_string(reason));
3416 			ret = EBUSY;
3417 			goto err;
3418 		}
3419 	}
3420 
3421 	rwi = kzalloc(sizeof(*rwi), GFP_ATOMIC);
3422 	if (!rwi) {
3423 		ret = ENOMEM;
3424 		goto err;
3425 	}
3426 	/* if we just received a transport event,
3427 	 * flush reset queue and process this reset
3428 	 */
3429 	if (adapter->force_reset_recovery)
3430 		flush_reset_queue(adapter);
3431 
3432 	rwi->reset_reason = reason;
3433 	list_add_tail(&rwi->list, &adapter->rwi_list);
3434 	netdev_dbg(adapter->netdev, "Scheduling reset (reason %s)\n",
3435 		   reset_reason_to_string(reason));
3436 	queue_work(system_long_wq, &adapter->ibmvnic_reset);
3437 
3438 	ret = 0;
3439 err:
3440 	/* ibmvnic_close() below can block, so drop the lock first */
3441 	spin_unlock_irqrestore(&adapter->rwi_lock, flags);
3442 
3443 	if (ret == ENOMEM)
3444 		ibmvnic_close(netdev);
3445 
3446 	return -ret;
3447 }
3448 
3449 static void ibmvnic_tx_timeout(struct net_device *dev, unsigned int txqueue)
3450 {
3451 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3452 
3453 	if (test_bit(0, &adapter->resetting)) {
3454 		netdev_err(adapter->netdev,
3455 			   "Adapter is resetting, skip timeout reset\n");
3456 		return;
3457 	}
3458 	/* No queuing up reset until at least 5 seconds (default watchdog val)
3459 	 * after last reset
3460 	 */
3461 	if (time_before(jiffies, (adapter->last_reset_time + dev->watchdog_timeo))) {
3462 		netdev_dbg(dev, "Not yet time to tx timeout.\n");
3463 		return;
3464 	}
3465 	ibmvnic_reset(adapter, VNIC_RESET_TIMEOUT);
3466 }
3467 
3468 static void remove_buff_from_pool(struct ibmvnic_adapter *adapter,
3469 				  struct ibmvnic_rx_buff *rx_buff)
3470 {
3471 	struct ibmvnic_rx_pool *pool = &adapter->rx_pool[rx_buff->pool_index];
3472 
3473 	rx_buff->skb = NULL;
3474 
3475 	pool->free_map[pool->next_alloc] = (int)(rx_buff - pool->rx_buff);
3476 	pool->next_alloc = (pool->next_alloc + 1) % pool->size;
3477 
3478 	atomic_dec(&pool->available);
3479 }
3480 
3481 static int ibmvnic_poll(struct napi_struct *napi, int budget)
3482 {
3483 	struct ibmvnic_sub_crq_queue *rx_scrq;
3484 	struct ibmvnic_adapter *adapter;
3485 	struct net_device *netdev;
3486 	int frames_processed;
3487 	int scrq_num;
3488 
3489 	netdev = napi->dev;
3490 	adapter = netdev_priv(netdev);
3491 	scrq_num = (int)(napi - adapter->napi);
3492 	frames_processed = 0;
3493 	rx_scrq = adapter->rx_scrq[scrq_num];
3494 
3495 restart_poll:
3496 	while (frames_processed < budget) {
3497 		struct sk_buff *skb;
3498 		struct ibmvnic_rx_buff *rx_buff;
3499 		union sub_crq *next;
3500 		u32 length;
3501 		u16 offset;
3502 		u8 flags = 0;
3503 
3504 		if (unlikely(test_bit(0, &adapter->resetting) &&
3505 			     adapter->reset_reason != VNIC_RESET_NON_FATAL)) {
3506 			enable_scrq_irq(adapter, rx_scrq);
3507 			napi_complete_done(napi, frames_processed);
3508 			return frames_processed;
3509 		}
3510 
3511 		if (!pending_scrq(adapter, rx_scrq))
3512 			break;
3513 		next = ibmvnic_next_scrq(adapter, rx_scrq);
3514 		rx_buff = (struct ibmvnic_rx_buff *)
3515 			  be64_to_cpu(next->rx_comp.correlator);
3516 		/* do error checking */
3517 		if (next->rx_comp.rc) {
3518 			netdev_dbg(netdev, "rx buffer returned with rc %x\n",
3519 				   be16_to_cpu(next->rx_comp.rc));
3520 			/* free the entry */
3521 			next->rx_comp.first = 0;
3522 			dev_kfree_skb_any(rx_buff->skb);
3523 			remove_buff_from_pool(adapter, rx_buff);
3524 			continue;
3525 		} else if (!rx_buff->skb) {
3526 			/* free the entry */
3527 			next->rx_comp.first = 0;
3528 			remove_buff_from_pool(adapter, rx_buff);
3529 			continue;
3530 		}
3531 
3532 		length = be32_to_cpu(next->rx_comp.len);
3533 		offset = be16_to_cpu(next->rx_comp.off_frame_data);
3534 		flags = next->rx_comp.flags;
3535 		skb = rx_buff->skb;
3536 		/* load long_term_buff before copying to skb */
3537 		dma_rmb();
3538 		skb_copy_to_linear_data(skb, rx_buff->data + offset,
3539 					length);
3540 
3541 		/* VLAN Header has been stripped by the system firmware and
3542 		 * needs to be inserted by the driver
3543 		 */
3544 		if (adapter->rx_vlan_header_insertion &&
3545 		    (flags & IBMVNIC_VLAN_STRIPPED))
3546 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
3547 					       ntohs(next->rx_comp.vlan_tci));
3548 
3549 		/* free the entry */
3550 		next->rx_comp.first = 0;
3551 		remove_buff_from_pool(adapter, rx_buff);
3552 
3553 		skb_put(skb, length);
3554 		skb->protocol = eth_type_trans(skb, netdev);
3555 		skb_record_rx_queue(skb, scrq_num);
3556 
3557 		if (flags & IBMVNIC_IP_CHKSUM_GOOD &&
3558 		    flags & IBMVNIC_TCP_UDP_CHKSUM_GOOD) {
3559 			skb->ip_summed = CHECKSUM_UNNECESSARY;
3560 		}
3561 
3562 		length = skb->len;
3563 		napi_gro_receive(napi, skb); /* send it up */
3564 		netdev->stats.rx_packets++;
3565 		netdev->stats.rx_bytes += length;
3566 		adapter->rx_stats_buffers[scrq_num].packets++;
3567 		adapter->rx_stats_buffers[scrq_num].bytes += length;
3568 		frames_processed++;
3569 	}
3570 
3571 	if (adapter->state != VNIC_CLOSING &&
3572 	    (atomic_read(&adapter->rx_pool[scrq_num].available) <
3573 	      adapter->req_rx_add_entries_per_subcrq / 2))
3574 		replenish_rx_pool(adapter, &adapter->rx_pool[scrq_num]);
3575 	if (frames_processed < budget) {
3576 		if (napi_complete_done(napi, frames_processed)) {
3577 			enable_scrq_irq(adapter, rx_scrq);
3578 			if (pending_scrq(adapter, rx_scrq)) {
3579 				if (napi_schedule(napi)) {
3580 					disable_scrq_irq(adapter, rx_scrq);
3581 					goto restart_poll;
3582 				}
3583 			}
3584 		}
3585 	}
3586 	return frames_processed;
3587 }
3588 
3589 static int wait_for_reset(struct ibmvnic_adapter *adapter)
3590 {
3591 	int rc, ret;
3592 
3593 	adapter->fallback.mtu = adapter->req_mtu;
3594 	adapter->fallback.rx_queues = adapter->req_rx_queues;
3595 	adapter->fallback.tx_queues = adapter->req_tx_queues;
3596 	adapter->fallback.rx_entries = adapter->req_rx_add_entries_per_subcrq;
3597 	adapter->fallback.tx_entries = adapter->req_tx_entries_per_subcrq;
3598 
3599 	reinit_completion(&adapter->reset_done);
3600 	adapter->wait_for_reset = true;
3601 	rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3602 
3603 	if (rc) {
3604 		ret = rc;
3605 		goto out;
3606 	}
3607 	rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done, 60000);
3608 	if (rc) {
3609 		ret = -ENODEV;
3610 		goto out;
3611 	}
3612 
3613 	ret = 0;
3614 	if (adapter->reset_done_rc) {
3615 		ret = -EIO;
3616 		adapter->desired.mtu = adapter->fallback.mtu;
3617 		adapter->desired.rx_queues = adapter->fallback.rx_queues;
3618 		adapter->desired.tx_queues = adapter->fallback.tx_queues;
3619 		adapter->desired.rx_entries = adapter->fallback.rx_entries;
3620 		adapter->desired.tx_entries = adapter->fallback.tx_entries;
3621 
3622 		reinit_completion(&adapter->reset_done);
3623 		adapter->wait_for_reset = true;
3624 		rc = ibmvnic_reset(adapter, VNIC_RESET_CHANGE_PARAM);
3625 		if (rc) {
3626 			ret = rc;
3627 			goto out;
3628 		}
3629 		rc = ibmvnic_wait_for_completion(adapter, &adapter->reset_done,
3630 						 60000);
3631 		if (rc) {
3632 			ret = -ENODEV;
3633 			goto out;
3634 		}
3635 	}
3636 out:
3637 	adapter->wait_for_reset = false;
3638 
3639 	return ret;
3640 }
3641 
3642 static int ibmvnic_change_mtu(struct net_device *netdev, int new_mtu)
3643 {
3644 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3645 
3646 	adapter->desired.mtu = new_mtu + ETH_HLEN;
3647 
3648 	return wait_for_reset(adapter);
3649 }
3650 
3651 static netdev_features_t ibmvnic_features_check(struct sk_buff *skb,
3652 						struct net_device *dev,
3653 						netdev_features_t features)
3654 {
3655 	/* Some backing hardware adapters can not
3656 	 * handle packets with a MSS less than 224
3657 	 * or with only one segment.
3658 	 */
3659 	if (skb_is_gso(skb)) {
3660 		if (skb_shinfo(skb)->gso_size < 224 ||
3661 		    skb_shinfo(skb)->gso_segs == 1)
3662 			features &= ~NETIF_F_GSO_MASK;
3663 	}
3664 
3665 	return features;
3666 }
3667 
3668 static const struct net_device_ops ibmvnic_netdev_ops = {
3669 	.ndo_open		= ibmvnic_open,
3670 	.ndo_stop		= ibmvnic_close,
3671 	.ndo_start_xmit		= ibmvnic_xmit,
3672 	.ndo_set_rx_mode	= ibmvnic_set_multi,
3673 	.ndo_set_mac_address	= ibmvnic_set_mac,
3674 	.ndo_validate_addr	= eth_validate_addr,
3675 	.ndo_tx_timeout		= ibmvnic_tx_timeout,
3676 	.ndo_change_mtu		= ibmvnic_change_mtu,
3677 	.ndo_features_check     = ibmvnic_features_check,
3678 };
3679 
3680 /* ethtool functions */
3681 
3682 static int ibmvnic_get_link_ksettings(struct net_device *netdev,
3683 				      struct ethtool_link_ksettings *cmd)
3684 {
3685 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3686 	int rc;
3687 
3688 	rc = send_query_phys_parms(adapter);
3689 	if (rc) {
3690 		adapter->speed = SPEED_UNKNOWN;
3691 		adapter->duplex = DUPLEX_UNKNOWN;
3692 	}
3693 	cmd->base.speed = adapter->speed;
3694 	cmd->base.duplex = adapter->duplex;
3695 	cmd->base.port = PORT_FIBRE;
3696 	cmd->base.phy_address = 0;
3697 	cmd->base.autoneg = AUTONEG_ENABLE;
3698 
3699 	return 0;
3700 }
3701 
3702 static void ibmvnic_get_drvinfo(struct net_device *netdev,
3703 				struct ethtool_drvinfo *info)
3704 {
3705 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3706 
3707 	strscpy(info->driver, ibmvnic_driver_name, sizeof(info->driver));
3708 	strscpy(info->version, IBMVNIC_DRIVER_VERSION, sizeof(info->version));
3709 	strscpy(info->fw_version, adapter->fw_version,
3710 		sizeof(info->fw_version));
3711 }
3712 
3713 static u32 ibmvnic_get_msglevel(struct net_device *netdev)
3714 {
3715 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3716 
3717 	return adapter->msg_enable;
3718 }
3719 
3720 static void ibmvnic_set_msglevel(struct net_device *netdev, u32 data)
3721 {
3722 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3723 
3724 	adapter->msg_enable = data;
3725 }
3726 
3727 static u32 ibmvnic_get_link(struct net_device *netdev)
3728 {
3729 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3730 
3731 	/* Don't need to send a query because we request a logical link up at
3732 	 * init and then we wait for link state indications
3733 	 */
3734 	return adapter->logical_link_state;
3735 }
3736 
3737 static void ibmvnic_get_ringparam(struct net_device *netdev,
3738 				  struct ethtool_ringparam *ring,
3739 				  struct kernel_ethtool_ringparam *kernel_ring,
3740 				  struct netlink_ext_ack *extack)
3741 {
3742 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3743 
3744 	ring->rx_max_pending = adapter->max_rx_add_entries_per_subcrq;
3745 	ring->tx_max_pending = adapter->max_tx_entries_per_subcrq;
3746 	ring->rx_mini_max_pending = 0;
3747 	ring->rx_jumbo_max_pending = 0;
3748 	ring->rx_pending = adapter->req_rx_add_entries_per_subcrq;
3749 	ring->tx_pending = adapter->req_tx_entries_per_subcrq;
3750 	ring->rx_mini_pending = 0;
3751 	ring->rx_jumbo_pending = 0;
3752 }
3753 
3754 static int ibmvnic_set_ringparam(struct net_device *netdev,
3755 				 struct ethtool_ringparam *ring,
3756 				 struct kernel_ethtool_ringparam *kernel_ring,
3757 				 struct netlink_ext_ack *extack)
3758 {
3759 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3760 
3761 	if (ring->rx_pending > adapter->max_rx_add_entries_per_subcrq  ||
3762 	    ring->tx_pending > adapter->max_tx_entries_per_subcrq) {
3763 		netdev_err(netdev, "Invalid request.\n");
3764 		netdev_err(netdev, "Max tx buffers = %llu\n",
3765 			   adapter->max_rx_add_entries_per_subcrq);
3766 		netdev_err(netdev, "Max rx buffers = %llu\n",
3767 			   adapter->max_tx_entries_per_subcrq);
3768 		return -EINVAL;
3769 	}
3770 
3771 	adapter->desired.rx_entries = ring->rx_pending;
3772 	adapter->desired.tx_entries = ring->tx_pending;
3773 
3774 	return wait_for_reset(adapter);
3775 }
3776 
3777 static void ibmvnic_get_channels(struct net_device *netdev,
3778 				 struct ethtool_channels *channels)
3779 {
3780 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3781 
3782 	channels->max_rx = adapter->max_rx_queues;
3783 	channels->max_tx = adapter->max_tx_queues;
3784 	channels->max_other = 0;
3785 	channels->max_combined = 0;
3786 	channels->rx_count = adapter->req_rx_queues;
3787 	channels->tx_count = adapter->req_tx_queues;
3788 	channels->other_count = 0;
3789 	channels->combined_count = 0;
3790 }
3791 
3792 static int ibmvnic_set_channels(struct net_device *netdev,
3793 				struct ethtool_channels *channels)
3794 {
3795 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
3796 
3797 	adapter->desired.rx_queues = channels->rx_count;
3798 	adapter->desired.tx_queues = channels->tx_count;
3799 
3800 	return wait_for_reset(adapter);
3801 }
3802 
3803 static void ibmvnic_get_strings(struct net_device *dev, u32 stringset, u8 *data)
3804 {
3805 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3806 	int i;
3807 
3808 	if (stringset != ETH_SS_STATS)
3809 		return;
3810 
3811 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3812 		ethtool_puts(&data, ibmvnic_stats[i].name);
3813 
3814 	for (i = 0; i < adapter->req_tx_queues; i++) {
3815 		ethtool_sprintf(&data, "tx%d_batched_packets", i);
3816 		ethtool_sprintf(&data, "tx%d_direct_packets", i);
3817 		ethtool_sprintf(&data, "tx%d_bytes", i);
3818 		ethtool_sprintf(&data, "tx%d_dropped_packets", i);
3819 	}
3820 
3821 	for (i = 0; i < adapter->req_rx_queues; i++) {
3822 		ethtool_sprintf(&data, "rx%d_packets", i);
3823 		ethtool_sprintf(&data, "rx%d_bytes", i);
3824 		ethtool_sprintf(&data, "rx%d_interrupts", i);
3825 	}
3826 }
3827 
3828 static int ibmvnic_get_sset_count(struct net_device *dev, int sset)
3829 {
3830 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3831 
3832 	switch (sset) {
3833 	case ETH_SS_STATS:
3834 		return ARRAY_SIZE(ibmvnic_stats) +
3835 		       adapter->req_tx_queues * NUM_TX_STATS +
3836 		       adapter->req_rx_queues * NUM_RX_STATS;
3837 	default:
3838 		return -EOPNOTSUPP;
3839 	}
3840 }
3841 
3842 static void ibmvnic_get_ethtool_stats(struct net_device *dev,
3843 				      struct ethtool_stats *stats, u64 *data)
3844 {
3845 	struct ibmvnic_adapter *adapter = netdev_priv(dev);
3846 	union ibmvnic_crq crq;
3847 	int i, j;
3848 	int rc;
3849 
3850 	memset(&crq, 0, sizeof(crq));
3851 	crq.request_statistics.first = IBMVNIC_CRQ_CMD;
3852 	crq.request_statistics.cmd = REQUEST_STATISTICS;
3853 	crq.request_statistics.ioba = cpu_to_be32(adapter->stats_token);
3854 	crq.request_statistics.len =
3855 	    cpu_to_be32(sizeof(struct ibmvnic_statistics));
3856 
3857 	/* Wait for data to be written */
3858 	reinit_completion(&adapter->stats_done);
3859 	rc = ibmvnic_send_crq(adapter, &crq);
3860 	if (rc)
3861 		return;
3862 	rc = ibmvnic_wait_for_completion(adapter, &adapter->stats_done, 10000);
3863 	if (rc)
3864 		return;
3865 
3866 	for (i = 0; i < ARRAY_SIZE(ibmvnic_stats); i++)
3867 		data[i] = be64_to_cpu(IBMVNIC_GET_STAT
3868 				      (adapter, ibmvnic_stats[i].offset));
3869 
3870 	for (j = 0; j < adapter->req_tx_queues; j++) {
3871 		data[i] = adapter->tx_stats_buffers[j].batched_packets;
3872 		i++;
3873 		data[i] = adapter->tx_stats_buffers[j].direct_packets;
3874 		i++;
3875 		data[i] = adapter->tx_stats_buffers[j].bytes;
3876 		i++;
3877 		data[i] = adapter->tx_stats_buffers[j].dropped_packets;
3878 		i++;
3879 	}
3880 
3881 	for (j = 0; j < adapter->req_rx_queues; j++) {
3882 		data[i] = adapter->rx_stats_buffers[j].packets;
3883 		i++;
3884 		data[i] = adapter->rx_stats_buffers[j].bytes;
3885 		i++;
3886 		data[i] = adapter->rx_stats_buffers[j].interrupts;
3887 		i++;
3888 	}
3889 }
3890 
3891 static const struct ethtool_ops ibmvnic_ethtool_ops = {
3892 	.get_drvinfo		= ibmvnic_get_drvinfo,
3893 	.get_msglevel		= ibmvnic_get_msglevel,
3894 	.set_msglevel		= ibmvnic_set_msglevel,
3895 	.get_link		= ibmvnic_get_link,
3896 	.get_ringparam		= ibmvnic_get_ringparam,
3897 	.set_ringparam		= ibmvnic_set_ringparam,
3898 	.get_channels		= ibmvnic_get_channels,
3899 	.set_channels		= ibmvnic_set_channels,
3900 	.get_strings            = ibmvnic_get_strings,
3901 	.get_sset_count         = ibmvnic_get_sset_count,
3902 	.get_ethtool_stats	= ibmvnic_get_ethtool_stats,
3903 	.get_link_ksettings	= ibmvnic_get_link_ksettings,
3904 };
3905 
3906 /* Routines for managing CRQs/sCRQs  */
3907 
3908 static int reset_one_sub_crq_queue(struct ibmvnic_adapter *adapter,
3909 				   struct ibmvnic_sub_crq_queue *scrq)
3910 {
3911 	int rc;
3912 
3913 	if (!scrq) {
3914 		netdev_dbg(adapter->netdev, "Invalid scrq reset.\n");
3915 		return -EINVAL;
3916 	}
3917 
3918 	if (scrq->irq) {
3919 		free_irq(scrq->irq, scrq);
3920 		irq_dispose_mapping(scrq->irq);
3921 		scrq->irq = 0;
3922 	}
3923 
3924 	if (scrq->msgs) {
3925 		memset(scrq->msgs, 0, 4 * PAGE_SIZE);
3926 		atomic_set(&scrq->used, 0);
3927 		scrq->cur = 0;
3928 		scrq->ind_buf.index = 0;
3929 	} else {
3930 		netdev_dbg(adapter->netdev, "Invalid scrq reset\n");
3931 		return -EINVAL;
3932 	}
3933 
3934 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
3935 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
3936 	return rc;
3937 }
3938 
3939 static int reset_sub_crq_queues(struct ibmvnic_adapter *adapter)
3940 {
3941 	int i, rc;
3942 
3943 	if (!adapter->tx_scrq || !adapter->rx_scrq)
3944 		return -EINVAL;
3945 
3946 	ibmvnic_clean_affinity(adapter);
3947 
3948 	for (i = 0; i < adapter->req_tx_queues; i++) {
3949 		netdev_dbg(adapter->netdev, "Re-setting tx_scrq[%d]\n", i);
3950 		rc = reset_one_sub_crq_queue(adapter, adapter->tx_scrq[i]);
3951 		if (rc)
3952 			return rc;
3953 	}
3954 
3955 	for (i = 0; i < adapter->req_rx_queues; i++) {
3956 		netdev_dbg(adapter->netdev, "Re-setting rx_scrq[%d]\n", i);
3957 		rc = reset_one_sub_crq_queue(adapter, adapter->rx_scrq[i]);
3958 		if (rc)
3959 			return rc;
3960 	}
3961 
3962 	return rc;
3963 }
3964 
3965 static void release_sub_crq_queue(struct ibmvnic_adapter *adapter,
3966 				  struct ibmvnic_sub_crq_queue *scrq,
3967 				  bool do_h_free)
3968 {
3969 	struct device *dev = &adapter->vdev->dev;
3970 	long rc;
3971 
3972 	netdev_dbg(adapter->netdev, "Releasing sub-CRQ\n");
3973 
3974 	if (do_h_free) {
3975 		/* Close the sub-crqs */
3976 		do {
3977 			rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
3978 						adapter->vdev->unit_address,
3979 						scrq->crq_num);
3980 		} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
3981 
3982 		if (rc) {
3983 			netdev_err(adapter->netdev,
3984 				   "Failed to release sub-CRQ %16lx, rc = %ld\n",
3985 				   scrq->crq_num, rc);
3986 		}
3987 	}
3988 
3989 	dma_free_coherent(dev,
3990 			  IBMVNIC_IND_ARR_SZ,
3991 			  scrq->ind_buf.indir_arr,
3992 			  scrq->ind_buf.indir_dma);
3993 
3994 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
3995 			 DMA_BIDIRECTIONAL);
3996 	free_pages((unsigned long)scrq->msgs, 2);
3997 	free_cpumask_var(scrq->affinity_mask);
3998 	kfree(scrq);
3999 }
4000 
4001 static struct ibmvnic_sub_crq_queue *init_sub_crq_queue(struct ibmvnic_adapter
4002 							*adapter)
4003 {
4004 	struct device *dev = &adapter->vdev->dev;
4005 	struct ibmvnic_sub_crq_queue *scrq;
4006 	int rc;
4007 
4008 	scrq = kzalloc(sizeof(*scrq), GFP_KERNEL);
4009 	if (!scrq)
4010 		return NULL;
4011 
4012 	scrq->msgs =
4013 		(union sub_crq *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 2);
4014 	if (!scrq->msgs) {
4015 		dev_warn(dev, "Couldn't allocate crq queue messages page\n");
4016 		goto zero_page_failed;
4017 	}
4018 	if (!zalloc_cpumask_var(&scrq->affinity_mask, GFP_KERNEL))
4019 		goto cpumask_alloc_failed;
4020 
4021 	scrq->msg_token = dma_map_single(dev, scrq->msgs, 4 * PAGE_SIZE,
4022 					 DMA_BIDIRECTIONAL);
4023 	if (dma_mapping_error(dev, scrq->msg_token)) {
4024 		dev_warn(dev, "Couldn't map crq queue messages page\n");
4025 		goto map_failed;
4026 	}
4027 
4028 	rc = h_reg_sub_crq(adapter->vdev->unit_address, scrq->msg_token,
4029 			   4 * PAGE_SIZE, &scrq->crq_num, &scrq->hw_irq);
4030 
4031 	if (rc == H_RESOURCE)
4032 		rc = ibmvnic_reset_crq(adapter);
4033 
4034 	if (rc == H_CLOSED) {
4035 		dev_warn(dev, "Partner adapter not ready, waiting.\n");
4036 	} else if (rc) {
4037 		dev_warn(dev, "Error %d registering sub-crq\n", rc);
4038 		goto reg_failed;
4039 	}
4040 
4041 	scrq->adapter = adapter;
4042 	scrq->size = 4 * PAGE_SIZE / sizeof(*scrq->msgs);
4043 	scrq->ind_buf.index = 0;
4044 
4045 	scrq->ind_buf.indir_arr =
4046 		dma_alloc_coherent(dev,
4047 				   IBMVNIC_IND_ARR_SZ,
4048 				   &scrq->ind_buf.indir_dma,
4049 				   GFP_KERNEL);
4050 
4051 	if (!scrq->ind_buf.indir_arr)
4052 		goto indir_failed;
4053 
4054 	spin_lock_init(&scrq->lock);
4055 
4056 	netdev_dbg(adapter->netdev,
4057 		   "sub-crq initialized, num %lx, hw_irq=%lx, irq=%x\n",
4058 		   scrq->crq_num, scrq->hw_irq, scrq->irq);
4059 
4060 	return scrq;
4061 
4062 indir_failed:
4063 	do {
4064 		rc = plpar_hcall_norets(H_FREE_SUB_CRQ,
4065 					adapter->vdev->unit_address,
4066 					scrq->crq_num);
4067 	} while (rc == H_BUSY || rc == H_IS_LONG_BUSY(rc));
4068 reg_failed:
4069 	dma_unmap_single(dev, scrq->msg_token, 4 * PAGE_SIZE,
4070 			 DMA_BIDIRECTIONAL);
4071 map_failed:
4072 	free_cpumask_var(scrq->affinity_mask);
4073 cpumask_alloc_failed:
4074 	free_pages((unsigned long)scrq->msgs, 2);
4075 zero_page_failed:
4076 	kfree(scrq);
4077 
4078 	return NULL;
4079 }
4080 
4081 static void release_sub_crqs(struct ibmvnic_adapter *adapter, bool do_h_free)
4082 {
4083 	int i;
4084 
4085 	ibmvnic_clean_affinity(adapter);
4086 	if (adapter->tx_scrq) {
4087 		for (i = 0; i < adapter->num_active_tx_scrqs; i++) {
4088 			if (!adapter->tx_scrq[i])
4089 				continue;
4090 
4091 			netdev_dbg(adapter->netdev, "Releasing tx_scrq[%d]\n",
4092 				   i);
4093 			ibmvnic_tx_scrq_clean_buffer(adapter, adapter->tx_scrq[i]);
4094 			if (adapter->tx_scrq[i]->irq) {
4095 				free_irq(adapter->tx_scrq[i]->irq,
4096 					 adapter->tx_scrq[i]);
4097 				irq_dispose_mapping(adapter->tx_scrq[i]->irq);
4098 				adapter->tx_scrq[i]->irq = 0;
4099 			}
4100 
4101 			release_sub_crq_queue(adapter, adapter->tx_scrq[i],
4102 					      do_h_free);
4103 		}
4104 
4105 		kfree(adapter->tx_scrq);
4106 		adapter->tx_scrq = NULL;
4107 		adapter->num_active_tx_scrqs = 0;
4108 	}
4109 
4110 	/* Clean any remaining outstanding SKBs
4111 	 * we freed the irq so we won't be hearing
4112 	 * from them
4113 	 */
4114 	clean_tx_pools(adapter);
4115 
4116 	if (adapter->rx_scrq) {
4117 		for (i = 0; i < adapter->num_active_rx_scrqs; i++) {
4118 			if (!adapter->rx_scrq[i])
4119 				continue;
4120 
4121 			netdev_dbg(adapter->netdev, "Releasing rx_scrq[%d]\n",
4122 				   i);
4123 			if (adapter->rx_scrq[i]->irq) {
4124 				free_irq(adapter->rx_scrq[i]->irq,
4125 					 adapter->rx_scrq[i]);
4126 				irq_dispose_mapping(adapter->rx_scrq[i]->irq);
4127 				adapter->rx_scrq[i]->irq = 0;
4128 			}
4129 
4130 			release_sub_crq_queue(adapter, adapter->rx_scrq[i],
4131 					      do_h_free);
4132 		}
4133 
4134 		kfree(adapter->rx_scrq);
4135 		adapter->rx_scrq = NULL;
4136 		adapter->num_active_rx_scrqs = 0;
4137 	}
4138 }
4139 
4140 static int disable_scrq_irq(struct ibmvnic_adapter *adapter,
4141 			    struct ibmvnic_sub_crq_queue *scrq)
4142 {
4143 	struct device *dev = &adapter->vdev->dev;
4144 	unsigned long rc;
4145 
4146 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4147 				H_DISABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4148 	if (rc)
4149 		dev_err(dev, "Couldn't disable scrq irq 0x%lx. rc=%ld\n",
4150 			scrq->hw_irq, rc);
4151 	return rc;
4152 }
4153 
4154 /* We can not use the IRQ chip EOI handler because that has the
4155  * unintended effect of changing the interrupt priority.
4156  */
4157 static void ibmvnic_xics_eoi(struct device *dev, struct ibmvnic_sub_crq_queue *scrq)
4158 {
4159 	u64 val = 0xff000000 | scrq->hw_irq;
4160 	unsigned long rc;
4161 
4162 	rc = plpar_hcall_norets(H_EOI, val);
4163 	if (rc)
4164 		dev_err(dev, "H_EOI FAILED irq 0x%llx. rc=%ld\n", val, rc);
4165 }
4166 
4167 /* Due to a firmware bug, the hypervisor can send an interrupt to a
4168  * transmit or receive queue just prior to a partition migration.
4169  * Force an EOI after migration.
4170  */
4171 static void ibmvnic_clear_pending_interrupt(struct device *dev,
4172 					    struct ibmvnic_sub_crq_queue *scrq)
4173 {
4174 	if (!xive_enabled())
4175 		ibmvnic_xics_eoi(dev, scrq);
4176 }
4177 
4178 static int enable_scrq_irq(struct ibmvnic_adapter *adapter,
4179 			   struct ibmvnic_sub_crq_queue *scrq)
4180 {
4181 	struct device *dev = &adapter->vdev->dev;
4182 	unsigned long rc;
4183 
4184 	if (scrq->hw_irq > 0x100000000ULL) {
4185 		dev_err(dev, "bad hw_irq = %lx\n", scrq->hw_irq);
4186 		return 1;
4187 	}
4188 
4189 	if (test_bit(0, &adapter->resetting) &&
4190 	    adapter->reset_reason == VNIC_RESET_MOBILITY) {
4191 		ibmvnic_clear_pending_interrupt(dev, scrq);
4192 	}
4193 
4194 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
4195 				H_ENABLE_VIO_INTERRUPT, scrq->hw_irq, 0, 0);
4196 	if (rc)
4197 		dev_err(dev, "Couldn't enable scrq irq 0x%lx. rc=%ld\n",
4198 			scrq->hw_irq, rc);
4199 	return rc;
4200 }
4201 
4202 static int ibmvnic_complete_tx(struct ibmvnic_adapter *adapter,
4203 			       struct ibmvnic_sub_crq_queue *scrq)
4204 {
4205 	struct device *dev = &adapter->vdev->dev;
4206 	int num_packets = 0, total_bytes = 0;
4207 	struct ibmvnic_tx_pool *tx_pool;
4208 	struct ibmvnic_tx_buff *txbuff;
4209 	struct netdev_queue *txq;
4210 	union sub_crq *next;
4211 	int index, i;
4212 
4213 restart_loop:
4214 	while (pending_scrq(adapter, scrq)) {
4215 		unsigned int pool = scrq->pool_index;
4216 		int num_entries = 0;
4217 		next = ibmvnic_next_scrq(adapter, scrq);
4218 		for (i = 0; i < next->tx_comp.num_comps; i++) {
4219 			index = be32_to_cpu(next->tx_comp.correlators[i]);
4220 			if (index & IBMVNIC_TSO_POOL_MASK) {
4221 				tx_pool = &adapter->tso_pool[pool];
4222 				index &= ~IBMVNIC_TSO_POOL_MASK;
4223 			} else {
4224 				tx_pool = &adapter->tx_pool[pool];
4225 			}
4226 
4227 			txbuff = &tx_pool->tx_buff[index];
4228 			num_packets++;
4229 			num_entries += txbuff->num_entries;
4230 			if (txbuff->skb) {
4231 				total_bytes += txbuff->skb->len;
4232 				if (next->tx_comp.rcs[i]) {
4233 					dev_err(dev, "tx error %x\n",
4234 						next->tx_comp.rcs[i]);
4235 					dev_kfree_skb_irq(txbuff->skb);
4236 				} else {
4237 					dev_consume_skb_irq(txbuff->skb);
4238 				}
4239 				txbuff->skb = NULL;
4240 			} else {
4241 				netdev_warn(adapter->netdev,
4242 					    "TX completion received with NULL socket buffer\n");
4243 			}
4244 			tx_pool->free_map[tx_pool->producer_index] = index;
4245 			tx_pool->producer_index =
4246 				(tx_pool->producer_index + 1) %
4247 					tx_pool->num_buffers;
4248 		}
4249 		/* remove tx_comp scrq*/
4250 		next->tx_comp.first = 0;
4251 
4252 
4253 		if (atomic_sub_return(num_entries, &scrq->used) <=
4254 		    (adapter->req_tx_entries_per_subcrq / 2) &&
4255 		    __netif_subqueue_stopped(adapter->netdev,
4256 					     scrq->pool_index)) {
4257 			rcu_read_lock();
4258 			if (adapter->tx_queues_active) {
4259 				netif_wake_subqueue(adapter->netdev,
4260 						    scrq->pool_index);
4261 				netdev_dbg(adapter->netdev,
4262 					   "Started queue %d\n",
4263 					   scrq->pool_index);
4264 			}
4265 			rcu_read_unlock();
4266 		}
4267 	}
4268 
4269 	enable_scrq_irq(adapter, scrq);
4270 
4271 	if (pending_scrq(adapter, scrq)) {
4272 		disable_scrq_irq(adapter, scrq);
4273 		goto restart_loop;
4274 	}
4275 
4276 	txq = netdev_get_tx_queue(adapter->netdev, scrq->pool_index);
4277 	netdev_tx_completed_queue(txq, num_packets, total_bytes);
4278 
4279 	return 0;
4280 }
4281 
4282 static irqreturn_t ibmvnic_interrupt_tx(int irq, void *instance)
4283 {
4284 	struct ibmvnic_sub_crq_queue *scrq = instance;
4285 	struct ibmvnic_adapter *adapter = scrq->adapter;
4286 
4287 	disable_scrq_irq(adapter, scrq);
4288 	ibmvnic_complete_tx(adapter, scrq);
4289 
4290 	return IRQ_HANDLED;
4291 }
4292 
4293 static irqreturn_t ibmvnic_interrupt_rx(int irq, void *instance)
4294 {
4295 	struct ibmvnic_sub_crq_queue *scrq = instance;
4296 	struct ibmvnic_adapter *adapter = scrq->adapter;
4297 
4298 	/* When booting a kdump kernel we can hit pending interrupts
4299 	 * prior to completing driver initialization.
4300 	 */
4301 	if (unlikely(adapter->state != VNIC_OPEN))
4302 		return IRQ_NONE;
4303 
4304 	adapter->rx_stats_buffers[scrq->scrq_num].interrupts++;
4305 
4306 	if (napi_schedule_prep(&adapter->napi[scrq->scrq_num])) {
4307 		disable_scrq_irq(adapter, scrq);
4308 		__napi_schedule(&adapter->napi[scrq->scrq_num]);
4309 	}
4310 
4311 	return IRQ_HANDLED;
4312 }
4313 
4314 static int init_sub_crq_irqs(struct ibmvnic_adapter *adapter)
4315 {
4316 	struct device *dev = &adapter->vdev->dev;
4317 	struct ibmvnic_sub_crq_queue *scrq;
4318 	int i = 0, j = 0;
4319 	int rc = 0;
4320 
4321 	for (i = 0; i < adapter->req_tx_queues; i++) {
4322 		netdev_dbg(adapter->netdev, "Initializing tx_scrq[%d] irq\n",
4323 			   i);
4324 		scrq = adapter->tx_scrq[i];
4325 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4326 
4327 		if (!scrq->irq) {
4328 			rc = -EINVAL;
4329 			dev_err(dev, "Error mapping irq\n");
4330 			goto req_tx_irq_failed;
4331 		}
4332 
4333 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-tx%d",
4334 			 adapter->vdev->unit_address, i);
4335 		rc = request_irq(scrq->irq, ibmvnic_interrupt_tx,
4336 				 0, scrq->name, scrq);
4337 
4338 		if (rc) {
4339 			dev_err(dev, "Couldn't register tx irq 0x%x. rc=%d\n",
4340 				scrq->irq, rc);
4341 			irq_dispose_mapping(scrq->irq);
4342 			goto req_tx_irq_failed;
4343 		}
4344 	}
4345 
4346 	for (i = 0; i < adapter->req_rx_queues; i++) {
4347 		netdev_dbg(adapter->netdev, "Initializing rx_scrq[%d] irq\n",
4348 			   i);
4349 		scrq = adapter->rx_scrq[i];
4350 		scrq->irq = irq_create_mapping(NULL, scrq->hw_irq);
4351 		if (!scrq->irq) {
4352 			rc = -EINVAL;
4353 			dev_err(dev, "Error mapping irq\n");
4354 			goto req_rx_irq_failed;
4355 		}
4356 		snprintf(scrq->name, sizeof(scrq->name), "ibmvnic-%x-rx%d",
4357 			 adapter->vdev->unit_address, i);
4358 		rc = request_irq(scrq->irq, ibmvnic_interrupt_rx,
4359 				 0, scrq->name, scrq);
4360 		if (rc) {
4361 			dev_err(dev, "Couldn't register rx irq 0x%x. rc=%d\n",
4362 				scrq->irq, rc);
4363 			irq_dispose_mapping(scrq->irq);
4364 			goto req_rx_irq_failed;
4365 		}
4366 	}
4367 
4368 	cpus_read_lock();
4369 	ibmvnic_set_affinity(adapter);
4370 	cpus_read_unlock();
4371 
4372 	return rc;
4373 
4374 req_rx_irq_failed:
4375 	for (j = 0; j < i; j++) {
4376 		free_irq(adapter->rx_scrq[j]->irq, adapter->rx_scrq[j]);
4377 		irq_dispose_mapping(adapter->rx_scrq[j]->irq);
4378 	}
4379 	i = adapter->req_tx_queues;
4380 req_tx_irq_failed:
4381 	for (j = 0; j < i; j++) {
4382 		free_irq(adapter->tx_scrq[j]->irq, adapter->tx_scrq[j]);
4383 		irq_dispose_mapping(adapter->tx_scrq[j]->irq);
4384 	}
4385 	release_sub_crqs(adapter, 1);
4386 	return rc;
4387 }
4388 
4389 static int init_sub_crqs(struct ibmvnic_adapter *adapter)
4390 {
4391 	struct device *dev = &adapter->vdev->dev;
4392 	struct ibmvnic_sub_crq_queue **allqueues;
4393 	int registered_queues = 0;
4394 	int total_queues;
4395 	int more = 0;
4396 	int i;
4397 
4398 	total_queues = adapter->req_tx_queues + adapter->req_rx_queues;
4399 
4400 	allqueues = kcalloc(total_queues, sizeof(*allqueues), GFP_KERNEL);
4401 	if (!allqueues)
4402 		return -ENOMEM;
4403 
4404 	for (i = 0; i < total_queues; i++) {
4405 		allqueues[i] = init_sub_crq_queue(adapter);
4406 		if (!allqueues[i]) {
4407 			dev_warn(dev, "Couldn't allocate all sub-crqs\n");
4408 			break;
4409 		}
4410 		registered_queues++;
4411 	}
4412 
4413 	/* Make sure we were able to register the minimum number of queues */
4414 	if (registered_queues <
4415 	    adapter->min_tx_queues + adapter->min_rx_queues) {
4416 		dev_err(dev, "Fatal: Couldn't init  min number of sub-crqs\n");
4417 		goto tx_failed;
4418 	}
4419 
4420 	/* Distribute the failed allocated queues*/
4421 	for (i = 0; i < total_queues - registered_queues + more ; i++) {
4422 		netdev_dbg(adapter->netdev, "Reducing number of queues\n");
4423 		switch (i % 3) {
4424 		case 0:
4425 			if (adapter->req_rx_queues > adapter->min_rx_queues)
4426 				adapter->req_rx_queues--;
4427 			else
4428 				more++;
4429 			break;
4430 		case 1:
4431 			if (adapter->req_tx_queues > adapter->min_tx_queues)
4432 				adapter->req_tx_queues--;
4433 			else
4434 				more++;
4435 			break;
4436 		}
4437 	}
4438 
4439 	adapter->tx_scrq = kcalloc(adapter->req_tx_queues,
4440 				   sizeof(*adapter->tx_scrq), GFP_KERNEL);
4441 	if (!adapter->tx_scrq)
4442 		goto tx_failed;
4443 
4444 	for (i = 0; i < adapter->req_tx_queues; i++) {
4445 		adapter->tx_scrq[i] = allqueues[i];
4446 		adapter->tx_scrq[i]->pool_index = i;
4447 		adapter->num_active_tx_scrqs++;
4448 	}
4449 
4450 	adapter->rx_scrq = kcalloc(adapter->req_rx_queues,
4451 				   sizeof(*adapter->rx_scrq), GFP_KERNEL);
4452 	if (!adapter->rx_scrq)
4453 		goto rx_failed;
4454 
4455 	for (i = 0; i < adapter->req_rx_queues; i++) {
4456 		adapter->rx_scrq[i] = allqueues[i + adapter->req_tx_queues];
4457 		adapter->rx_scrq[i]->scrq_num = i;
4458 		adapter->num_active_rx_scrqs++;
4459 	}
4460 
4461 	kfree(allqueues);
4462 	return 0;
4463 
4464 rx_failed:
4465 	kfree(adapter->tx_scrq);
4466 	adapter->tx_scrq = NULL;
4467 tx_failed:
4468 	for (i = 0; i < registered_queues; i++)
4469 		release_sub_crq_queue(adapter, allqueues[i], 1);
4470 	kfree(allqueues);
4471 	return -ENOMEM;
4472 }
4473 
4474 static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
4475 {
4476 	struct device *dev = &adapter->vdev->dev;
4477 	union ibmvnic_crq crq;
4478 	int max_entries;
4479 	int cap_reqs;
4480 
4481 	/* We send out 6 or 7 REQUEST_CAPABILITY CRQs below (depending on
4482 	 * the PROMISC flag). Initialize this count upfront. When the tasklet
4483 	 * receives a response to all of these, it will send the next protocol
4484 	 * message (QUERY_IP_OFFLOAD).
4485 	 */
4486 	if (!(adapter->netdev->flags & IFF_PROMISC) ||
4487 	    adapter->promisc_supported)
4488 		cap_reqs = 7;
4489 	else
4490 		cap_reqs = 6;
4491 
4492 	if (!retry) {
4493 		/* Sub-CRQ entries are 32 byte long */
4494 		int entries_page = 4 * PAGE_SIZE / (sizeof(u64) * 4);
4495 
4496 		atomic_set(&adapter->running_cap_crqs, cap_reqs);
4497 
4498 		if (adapter->min_tx_entries_per_subcrq > entries_page ||
4499 		    adapter->min_rx_add_entries_per_subcrq > entries_page) {
4500 			dev_err(dev, "Fatal, invalid entries per sub-crq\n");
4501 			return;
4502 		}
4503 
4504 		if (adapter->desired.mtu)
4505 			adapter->req_mtu = adapter->desired.mtu;
4506 		else
4507 			adapter->req_mtu = adapter->netdev->mtu + ETH_HLEN;
4508 
4509 		if (!adapter->desired.tx_entries)
4510 			adapter->desired.tx_entries =
4511 					adapter->max_tx_entries_per_subcrq;
4512 		if (!adapter->desired.rx_entries)
4513 			adapter->desired.rx_entries =
4514 					adapter->max_rx_add_entries_per_subcrq;
4515 
4516 		max_entries = IBMVNIC_LTB_SET_SIZE /
4517 			      (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
4518 
4519 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4520 			adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) {
4521 			adapter->desired.tx_entries = max_entries;
4522 		}
4523 
4524 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
4525 			adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) {
4526 			adapter->desired.rx_entries = max_entries;
4527 		}
4528 
4529 		if (adapter->desired.tx_entries)
4530 			adapter->req_tx_entries_per_subcrq =
4531 					adapter->desired.tx_entries;
4532 		else
4533 			adapter->req_tx_entries_per_subcrq =
4534 					adapter->max_tx_entries_per_subcrq;
4535 
4536 		if (adapter->desired.rx_entries)
4537 			adapter->req_rx_add_entries_per_subcrq =
4538 					adapter->desired.rx_entries;
4539 		else
4540 			adapter->req_rx_add_entries_per_subcrq =
4541 					adapter->max_rx_add_entries_per_subcrq;
4542 
4543 		if (adapter->desired.tx_queues)
4544 			adapter->req_tx_queues =
4545 					adapter->desired.tx_queues;
4546 		else
4547 			adapter->req_tx_queues =
4548 					adapter->opt_tx_comp_sub_queues;
4549 
4550 		if (adapter->desired.rx_queues)
4551 			adapter->req_rx_queues =
4552 					adapter->desired.rx_queues;
4553 		else
4554 			adapter->req_rx_queues =
4555 					adapter->opt_rx_comp_queues;
4556 
4557 		adapter->req_rx_add_queues = adapter->max_rx_add_queues;
4558 	} else {
4559 		atomic_add(cap_reqs, &adapter->running_cap_crqs);
4560 	}
4561 	memset(&crq, 0, sizeof(crq));
4562 	crq.request_capability.first = IBMVNIC_CRQ_CMD;
4563 	crq.request_capability.cmd = REQUEST_CAPABILITY;
4564 
4565 	crq.request_capability.capability = cpu_to_be16(REQ_TX_QUEUES);
4566 	crq.request_capability.number = cpu_to_be64(adapter->req_tx_queues);
4567 	cap_reqs--;
4568 	ibmvnic_send_crq(adapter, &crq);
4569 
4570 	crq.request_capability.capability = cpu_to_be16(REQ_RX_QUEUES);
4571 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_queues);
4572 	cap_reqs--;
4573 	ibmvnic_send_crq(adapter, &crq);
4574 
4575 	crq.request_capability.capability = cpu_to_be16(REQ_RX_ADD_QUEUES);
4576 	crq.request_capability.number = cpu_to_be64(adapter->req_rx_add_queues);
4577 	cap_reqs--;
4578 	ibmvnic_send_crq(adapter, &crq);
4579 
4580 	crq.request_capability.capability =
4581 	    cpu_to_be16(REQ_TX_ENTRIES_PER_SUBCRQ);
4582 	crq.request_capability.number =
4583 	    cpu_to_be64(adapter->req_tx_entries_per_subcrq);
4584 	cap_reqs--;
4585 	ibmvnic_send_crq(adapter, &crq);
4586 
4587 	crq.request_capability.capability =
4588 	    cpu_to_be16(REQ_RX_ADD_ENTRIES_PER_SUBCRQ);
4589 	crq.request_capability.number =
4590 	    cpu_to_be64(adapter->req_rx_add_entries_per_subcrq);
4591 	cap_reqs--;
4592 	ibmvnic_send_crq(adapter, &crq);
4593 
4594 	crq.request_capability.capability = cpu_to_be16(REQ_MTU);
4595 	crq.request_capability.number = cpu_to_be64(adapter->req_mtu);
4596 	cap_reqs--;
4597 	ibmvnic_send_crq(adapter, &crq);
4598 
4599 	if (adapter->netdev->flags & IFF_PROMISC) {
4600 		if (adapter->promisc_supported) {
4601 			crq.request_capability.capability =
4602 			    cpu_to_be16(PROMISC_REQUESTED);
4603 			crq.request_capability.number = cpu_to_be64(1);
4604 			cap_reqs--;
4605 			ibmvnic_send_crq(adapter, &crq);
4606 		}
4607 	} else {
4608 		crq.request_capability.capability =
4609 		    cpu_to_be16(PROMISC_REQUESTED);
4610 		crq.request_capability.number = cpu_to_be64(0);
4611 		cap_reqs--;
4612 		ibmvnic_send_crq(adapter, &crq);
4613 	}
4614 
4615 	/* Keep at end to catch any discrepancy between expected and actual
4616 	 * CRQs sent.
4617 	 */
4618 	WARN_ON(cap_reqs != 0);
4619 }
4620 
4621 static int pending_scrq(struct ibmvnic_adapter *adapter,
4622 			struct ibmvnic_sub_crq_queue *scrq)
4623 {
4624 	union sub_crq *entry = &scrq->msgs[scrq->cur];
4625 	int rc;
4626 
4627 	rc = !!(entry->generic.first & IBMVNIC_CRQ_CMD_RSP);
4628 
4629 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4630 	 * contents of the SCRQ descriptor
4631 	 */
4632 	dma_rmb();
4633 
4634 	return rc;
4635 }
4636 
4637 static union sub_crq *ibmvnic_next_scrq(struct ibmvnic_adapter *adapter,
4638 					struct ibmvnic_sub_crq_queue *scrq)
4639 {
4640 	union sub_crq *entry;
4641 	unsigned long flags;
4642 
4643 	spin_lock_irqsave(&scrq->lock, flags);
4644 	entry = &scrq->msgs[scrq->cur];
4645 	if (entry->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4646 		if (++scrq->cur == scrq->size)
4647 			scrq->cur = 0;
4648 	} else {
4649 		entry = NULL;
4650 	}
4651 	spin_unlock_irqrestore(&scrq->lock, flags);
4652 
4653 	/* Ensure that the SCRQ valid flag is loaded prior to loading the
4654 	 * contents of the SCRQ descriptor
4655 	 */
4656 	dma_rmb();
4657 
4658 	return entry;
4659 }
4660 
4661 static union ibmvnic_crq *ibmvnic_next_crq(struct ibmvnic_adapter *adapter)
4662 {
4663 	struct ibmvnic_crq_queue *queue = &adapter->crq;
4664 	union ibmvnic_crq *crq;
4665 
4666 	crq = &queue->msgs[queue->cur];
4667 	if (crq->generic.first & IBMVNIC_CRQ_CMD_RSP) {
4668 		if (++queue->cur == queue->size)
4669 			queue->cur = 0;
4670 	} else {
4671 		crq = NULL;
4672 	}
4673 
4674 	return crq;
4675 }
4676 
4677 static void print_subcrq_error(struct device *dev, int rc, const char *func)
4678 {
4679 	switch (rc) {
4680 	case H_PARAMETER:
4681 		dev_warn_ratelimited(dev,
4682 				     "%s failed: Send request is malformed or adapter failover pending. (rc=%d)\n",
4683 				     func, rc);
4684 		break;
4685 	case H_CLOSED:
4686 		dev_warn_ratelimited(dev,
4687 				     "%s failed: Backing queue closed. Adapter is down or failover pending. (rc=%d)\n",
4688 				     func, rc);
4689 		break;
4690 	default:
4691 		dev_err_ratelimited(dev, "%s failed: (rc=%d)\n", func, rc);
4692 		break;
4693 	}
4694 }
4695 
4696 static int send_subcrq_indirect(struct ibmvnic_adapter *adapter,
4697 				u64 remote_handle, u64 ioba, u64 num_entries)
4698 {
4699 	unsigned int ua = adapter->vdev->unit_address;
4700 	struct device *dev = &adapter->vdev->dev;
4701 	int rc;
4702 
4703 	/* Make sure the hypervisor sees the complete request */
4704 	dma_wmb();
4705 	rc = plpar_hcall_norets(H_SEND_SUB_CRQ_INDIRECT, ua,
4706 				cpu_to_be64(remote_handle),
4707 				ioba, num_entries);
4708 
4709 	if (rc)
4710 		print_subcrq_error(dev, rc, __func__);
4711 
4712 	return rc;
4713 }
4714 
4715 static int ibmvnic_send_crq(struct ibmvnic_adapter *adapter,
4716 			    union ibmvnic_crq *crq)
4717 {
4718 	unsigned int ua = adapter->vdev->unit_address;
4719 	struct device *dev = &adapter->vdev->dev;
4720 	u64 *u64_crq = (u64 *)crq;
4721 	int rc;
4722 
4723 	netdev_dbg(adapter->netdev, "Sending CRQ: %016lx %016lx\n",
4724 		   (unsigned long)cpu_to_be64(u64_crq[0]),
4725 		   (unsigned long)cpu_to_be64(u64_crq[1]));
4726 
4727 	if (!adapter->crq.active &&
4728 	    crq->generic.first != IBMVNIC_CRQ_INIT_CMD) {
4729 		dev_warn(dev, "Invalid request detected while CRQ is inactive, possible device state change during reset\n");
4730 		return -EINVAL;
4731 	}
4732 
4733 	/* Make sure the hypervisor sees the complete request */
4734 	dma_wmb();
4735 
4736 	rc = plpar_hcall_norets(H_SEND_CRQ, ua,
4737 				cpu_to_be64(u64_crq[0]),
4738 				cpu_to_be64(u64_crq[1]));
4739 
4740 	if (rc) {
4741 		if (rc == H_CLOSED) {
4742 			dev_warn(dev, "CRQ Queue closed\n");
4743 			/* do not reset, report the fail, wait for passive init from server */
4744 		}
4745 
4746 		dev_warn(dev, "Send error (rc=%d)\n", rc);
4747 	}
4748 
4749 	return rc;
4750 }
4751 
4752 static int ibmvnic_send_crq_init(struct ibmvnic_adapter *adapter)
4753 {
4754 	struct device *dev = &adapter->vdev->dev;
4755 	union ibmvnic_crq crq;
4756 	int retries = 100;
4757 	int rc;
4758 
4759 	memset(&crq, 0, sizeof(crq));
4760 	crq.generic.first = IBMVNIC_CRQ_INIT_CMD;
4761 	crq.generic.cmd = IBMVNIC_CRQ_INIT;
4762 	netdev_dbg(adapter->netdev, "Sending CRQ init\n");
4763 
4764 	do {
4765 		rc = ibmvnic_send_crq(adapter, &crq);
4766 		if (rc != H_CLOSED)
4767 			break;
4768 		retries--;
4769 		msleep(50);
4770 
4771 	} while (retries > 0);
4772 
4773 	if (rc) {
4774 		dev_err(dev, "Failed to send init request, rc = %d\n", rc);
4775 		return rc;
4776 	}
4777 
4778 	return 0;
4779 }
4780 
4781 struct vnic_login_client_data {
4782 	u8	type;
4783 	__be16	len;
4784 	char	name[];
4785 } __packed;
4786 
4787 static int vnic_client_data_len(struct ibmvnic_adapter *adapter)
4788 {
4789 	int len;
4790 
4791 	/* Calculate the amount of buffer space needed for the
4792 	 * vnic client data in the login buffer. There are four entries,
4793 	 * OS name, LPAR name, device name, and a null last entry.
4794 	 */
4795 	len = 4 * sizeof(struct vnic_login_client_data);
4796 	len += 6; /* "Linux" plus NULL */
4797 	len += strlen(utsname()->nodename) + 1;
4798 	len += strlen(adapter->netdev->name) + 1;
4799 
4800 	return len;
4801 }
4802 
4803 static void vnic_add_client_data(struct ibmvnic_adapter *adapter,
4804 				 struct vnic_login_client_data *vlcd)
4805 {
4806 	const char *os_name = "Linux";
4807 	int len;
4808 
4809 	/* Type 1 - LPAR OS */
4810 	vlcd->type = 1;
4811 	len = strlen(os_name) + 1;
4812 	vlcd->len = cpu_to_be16(len);
4813 	strscpy(vlcd->name, os_name, len);
4814 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4815 
4816 	/* Type 2 - LPAR name */
4817 	vlcd->type = 2;
4818 	len = strlen(utsname()->nodename) + 1;
4819 	vlcd->len = cpu_to_be16(len);
4820 	strscpy(vlcd->name, utsname()->nodename, len);
4821 	vlcd = (struct vnic_login_client_data *)(vlcd->name + len);
4822 
4823 	/* Type 3 - device name */
4824 	vlcd->type = 3;
4825 	len = strlen(adapter->netdev->name) + 1;
4826 	vlcd->len = cpu_to_be16(len);
4827 	strscpy(vlcd->name, adapter->netdev->name, len);
4828 }
4829 
4830 static int send_login(struct ibmvnic_adapter *adapter)
4831 {
4832 	struct ibmvnic_login_rsp_buffer *login_rsp_buffer;
4833 	struct ibmvnic_login_buffer *login_buffer;
4834 	struct device *dev = &adapter->vdev->dev;
4835 	struct vnic_login_client_data *vlcd;
4836 	dma_addr_t rsp_buffer_token;
4837 	dma_addr_t buffer_token;
4838 	size_t rsp_buffer_size;
4839 	union ibmvnic_crq crq;
4840 	int client_data_len;
4841 	size_t buffer_size;
4842 	__be64 *tx_list_p;
4843 	__be64 *rx_list_p;
4844 	int rc;
4845 	int i;
4846 
4847 	if (!adapter->tx_scrq || !adapter->rx_scrq) {
4848 		netdev_err(adapter->netdev,
4849 			   "RX or TX queues are not allocated, device login failed\n");
4850 		return -ENOMEM;
4851 	}
4852 
4853 	release_login_buffer(adapter);
4854 	release_login_rsp_buffer(adapter);
4855 
4856 	client_data_len = vnic_client_data_len(adapter);
4857 
4858 	buffer_size =
4859 	    sizeof(struct ibmvnic_login_buffer) +
4860 	    sizeof(u64) * (adapter->req_tx_queues + adapter->req_rx_queues) +
4861 	    client_data_len;
4862 
4863 	login_buffer = kzalloc(buffer_size, GFP_ATOMIC);
4864 	if (!login_buffer)
4865 		goto buf_alloc_failed;
4866 
4867 	buffer_token = dma_map_single(dev, login_buffer, buffer_size,
4868 				      DMA_TO_DEVICE);
4869 	if (dma_mapping_error(dev, buffer_token)) {
4870 		dev_err(dev, "Couldn't map login buffer\n");
4871 		goto buf_map_failed;
4872 	}
4873 
4874 	rsp_buffer_size = sizeof(struct ibmvnic_login_rsp_buffer) +
4875 			  sizeof(u64) * adapter->req_tx_queues +
4876 			  sizeof(u64) * adapter->req_rx_queues +
4877 			  sizeof(u64) * adapter->req_rx_queues +
4878 			  sizeof(u8) * IBMVNIC_TX_DESC_VERSIONS;
4879 
4880 	login_rsp_buffer = kmalloc(rsp_buffer_size, GFP_ATOMIC);
4881 	if (!login_rsp_buffer)
4882 		goto buf_rsp_alloc_failed;
4883 
4884 	rsp_buffer_token = dma_map_single(dev, login_rsp_buffer,
4885 					  rsp_buffer_size, DMA_FROM_DEVICE);
4886 	if (dma_mapping_error(dev, rsp_buffer_token)) {
4887 		dev_err(dev, "Couldn't map login rsp buffer\n");
4888 		goto buf_rsp_map_failed;
4889 	}
4890 
4891 	adapter->login_buf = login_buffer;
4892 	adapter->login_buf_token = buffer_token;
4893 	adapter->login_buf_sz = buffer_size;
4894 	adapter->login_rsp_buf = login_rsp_buffer;
4895 	adapter->login_rsp_buf_token = rsp_buffer_token;
4896 	adapter->login_rsp_buf_sz = rsp_buffer_size;
4897 
4898 	login_buffer->len = cpu_to_be32(buffer_size);
4899 	login_buffer->version = cpu_to_be32(INITIAL_VERSION_LB);
4900 	login_buffer->num_txcomp_subcrqs = cpu_to_be32(adapter->req_tx_queues);
4901 	login_buffer->off_txcomp_subcrqs =
4902 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer));
4903 	login_buffer->num_rxcomp_subcrqs = cpu_to_be32(adapter->req_rx_queues);
4904 	login_buffer->off_rxcomp_subcrqs =
4905 	    cpu_to_be32(sizeof(struct ibmvnic_login_buffer) +
4906 			sizeof(u64) * adapter->req_tx_queues);
4907 	login_buffer->login_rsp_ioba = cpu_to_be32(rsp_buffer_token);
4908 	login_buffer->login_rsp_len = cpu_to_be32(rsp_buffer_size);
4909 
4910 	tx_list_p = (__be64 *)((char *)login_buffer +
4911 				      sizeof(struct ibmvnic_login_buffer));
4912 	rx_list_p = (__be64 *)((char *)login_buffer +
4913 				      sizeof(struct ibmvnic_login_buffer) +
4914 				      sizeof(u64) * adapter->req_tx_queues);
4915 
4916 	for (i = 0; i < adapter->req_tx_queues; i++) {
4917 		if (adapter->tx_scrq[i]) {
4918 			tx_list_p[i] =
4919 				cpu_to_be64(adapter->tx_scrq[i]->crq_num);
4920 		}
4921 	}
4922 
4923 	for (i = 0; i < adapter->req_rx_queues; i++) {
4924 		if (adapter->rx_scrq[i]) {
4925 			rx_list_p[i] =
4926 				cpu_to_be64(adapter->rx_scrq[i]->crq_num);
4927 		}
4928 	}
4929 
4930 	/* Insert vNIC login client data */
4931 	vlcd = (struct vnic_login_client_data *)
4932 		((char *)rx_list_p + (sizeof(u64) * adapter->req_rx_queues));
4933 	login_buffer->client_data_offset =
4934 			cpu_to_be32((char *)vlcd - (char *)login_buffer);
4935 	login_buffer->client_data_len = cpu_to_be32(client_data_len);
4936 
4937 	vnic_add_client_data(adapter, vlcd);
4938 
4939 	netdev_dbg(adapter->netdev, "Login Buffer:\n");
4940 	for (i = 0; i < (adapter->login_buf_sz - 1) / 8 + 1; i++) {
4941 		netdev_dbg(adapter->netdev, "%016lx\n",
4942 			   ((unsigned long *)(adapter->login_buf))[i]);
4943 	}
4944 
4945 	memset(&crq, 0, sizeof(crq));
4946 	crq.login.first = IBMVNIC_CRQ_CMD;
4947 	crq.login.cmd = LOGIN;
4948 	crq.login.ioba = cpu_to_be32(buffer_token);
4949 	crq.login.len = cpu_to_be32(buffer_size);
4950 
4951 	adapter->login_pending = true;
4952 	rc = ibmvnic_send_crq(adapter, &crq);
4953 	if (rc) {
4954 		adapter->login_pending = false;
4955 		netdev_err(adapter->netdev, "Failed to send login, rc=%d\n", rc);
4956 		goto buf_send_failed;
4957 	}
4958 
4959 	return 0;
4960 
4961 buf_send_failed:
4962 	dma_unmap_single(dev, rsp_buffer_token, rsp_buffer_size,
4963 			 DMA_FROM_DEVICE);
4964 buf_rsp_map_failed:
4965 	kfree(login_rsp_buffer);
4966 	adapter->login_rsp_buf = NULL;
4967 buf_rsp_alloc_failed:
4968 	dma_unmap_single(dev, buffer_token, buffer_size, DMA_TO_DEVICE);
4969 buf_map_failed:
4970 	kfree(login_buffer);
4971 	adapter->login_buf = NULL;
4972 buf_alloc_failed:
4973 	return -ENOMEM;
4974 }
4975 
4976 static int send_request_map(struct ibmvnic_adapter *adapter, dma_addr_t addr,
4977 			    u32 len, u8 map_id)
4978 {
4979 	union ibmvnic_crq crq;
4980 
4981 	memset(&crq, 0, sizeof(crq));
4982 	crq.request_map.first = IBMVNIC_CRQ_CMD;
4983 	crq.request_map.cmd = REQUEST_MAP;
4984 	crq.request_map.map_id = map_id;
4985 	crq.request_map.ioba = cpu_to_be32(addr);
4986 	crq.request_map.len = cpu_to_be32(len);
4987 	return ibmvnic_send_crq(adapter, &crq);
4988 }
4989 
4990 static int send_request_unmap(struct ibmvnic_adapter *adapter, u8 map_id)
4991 {
4992 	union ibmvnic_crq crq;
4993 
4994 	memset(&crq, 0, sizeof(crq));
4995 	crq.request_unmap.first = IBMVNIC_CRQ_CMD;
4996 	crq.request_unmap.cmd = REQUEST_UNMAP;
4997 	crq.request_unmap.map_id = map_id;
4998 	return ibmvnic_send_crq(adapter, &crq);
4999 }
5000 
5001 static void send_query_map(struct ibmvnic_adapter *adapter)
5002 {
5003 	union ibmvnic_crq crq;
5004 
5005 	memset(&crq, 0, sizeof(crq));
5006 	crq.query_map.first = IBMVNIC_CRQ_CMD;
5007 	crq.query_map.cmd = QUERY_MAP;
5008 	ibmvnic_send_crq(adapter, &crq);
5009 }
5010 
5011 /* Send a series of CRQs requesting various capabilities of the VNIC server */
5012 static void send_query_cap(struct ibmvnic_adapter *adapter)
5013 {
5014 	union ibmvnic_crq crq;
5015 	int cap_reqs;
5016 
5017 	/* We send out 25 QUERY_CAPABILITY CRQs below.  Initialize this count
5018 	 * upfront. When the tasklet receives a response to all of these, it
5019 	 * can send out the next protocol messaage (REQUEST_CAPABILITY).
5020 	 */
5021 	cap_reqs = 25;
5022 
5023 	atomic_set(&adapter->running_cap_crqs, cap_reqs);
5024 
5025 	memset(&crq, 0, sizeof(crq));
5026 	crq.query_capability.first = IBMVNIC_CRQ_CMD;
5027 	crq.query_capability.cmd = QUERY_CAPABILITY;
5028 
5029 	crq.query_capability.capability = cpu_to_be16(MIN_TX_QUEUES);
5030 	ibmvnic_send_crq(adapter, &crq);
5031 	cap_reqs--;
5032 
5033 	crq.query_capability.capability = cpu_to_be16(MIN_RX_QUEUES);
5034 	ibmvnic_send_crq(adapter, &crq);
5035 	cap_reqs--;
5036 
5037 	crq.query_capability.capability = cpu_to_be16(MIN_RX_ADD_QUEUES);
5038 	ibmvnic_send_crq(adapter, &crq);
5039 	cap_reqs--;
5040 
5041 	crq.query_capability.capability = cpu_to_be16(MAX_TX_QUEUES);
5042 	ibmvnic_send_crq(adapter, &crq);
5043 	cap_reqs--;
5044 
5045 	crq.query_capability.capability = cpu_to_be16(MAX_RX_QUEUES);
5046 	ibmvnic_send_crq(adapter, &crq);
5047 	cap_reqs--;
5048 
5049 	crq.query_capability.capability = cpu_to_be16(MAX_RX_ADD_QUEUES);
5050 	ibmvnic_send_crq(adapter, &crq);
5051 	cap_reqs--;
5052 
5053 	crq.query_capability.capability =
5054 	    cpu_to_be16(MIN_TX_ENTRIES_PER_SUBCRQ);
5055 	ibmvnic_send_crq(adapter, &crq);
5056 	cap_reqs--;
5057 
5058 	crq.query_capability.capability =
5059 	    cpu_to_be16(MIN_RX_ADD_ENTRIES_PER_SUBCRQ);
5060 	ibmvnic_send_crq(adapter, &crq);
5061 	cap_reqs--;
5062 
5063 	crq.query_capability.capability =
5064 	    cpu_to_be16(MAX_TX_ENTRIES_PER_SUBCRQ);
5065 	ibmvnic_send_crq(adapter, &crq);
5066 	cap_reqs--;
5067 
5068 	crq.query_capability.capability =
5069 	    cpu_to_be16(MAX_RX_ADD_ENTRIES_PER_SUBCRQ);
5070 	ibmvnic_send_crq(adapter, &crq);
5071 	cap_reqs--;
5072 
5073 	crq.query_capability.capability = cpu_to_be16(TCP_IP_OFFLOAD);
5074 	ibmvnic_send_crq(adapter, &crq);
5075 	cap_reqs--;
5076 
5077 	crq.query_capability.capability = cpu_to_be16(PROMISC_SUPPORTED);
5078 	ibmvnic_send_crq(adapter, &crq);
5079 	cap_reqs--;
5080 
5081 	crq.query_capability.capability = cpu_to_be16(MIN_MTU);
5082 	ibmvnic_send_crq(adapter, &crq);
5083 	cap_reqs--;
5084 
5085 	crq.query_capability.capability = cpu_to_be16(MAX_MTU);
5086 	ibmvnic_send_crq(adapter, &crq);
5087 	cap_reqs--;
5088 
5089 	crq.query_capability.capability = cpu_to_be16(MAX_MULTICAST_FILTERS);
5090 	ibmvnic_send_crq(adapter, &crq);
5091 	cap_reqs--;
5092 
5093 	crq.query_capability.capability = cpu_to_be16(VLAN_HEADER_INSERTION);
5094 	ibmvnic_send_crq(adapter, &crq);
5095 	cap_reqs--;
5096 
5097 	crq.query_capability.capability = cpu_to_be16(RX_VLAN_HEADER_INSERTION);
5098 	ibmvnic_send_crq(adapter, &crq);
5099 	cap_reqs--;
5100 
5101 	crq.query_capability.capability = cpu_to_be16(MAX_TX_SG_ENTRIES);
5102 	ibmvnic_send_crq(adapter, &crq);
5103 	cap_reqs--;
5104 
5105 	crq.query_capability.capability = cpu_to_be16(RX_SG_SUPPORTED);
5106 	ibmvnic_send_crq(adapter, &crq);
5107 	cap_reqs--;
5108 
5109 	crq.query_capability.capability = cpu_to_be16(OPT_TX_COMP_SUB_QUEUES);
5110 	ibmvnic_send_crq(adapter, &crq);
5111 	cap_reqs--;
5112 
5113 	crq.query_capability.capability = cpu_to_be16(OPT_RX_COMP_QUEUES);
5114 	ibmvnic_send_crq(adapter, &crq);
5115 	cap_reqs--;
5116 
5117 	crq.query_capability.capability =
5118 			cpu_to_be16(OPT_RX_BUFADD_Q_PER_RX_COMP_Q);
5119 	ibmvnic_send_crq(adapter, &crq);
5120 	cap_reqs--;
5121 
5122 	crq.query_capability.capability =
5123 			cpu_to_be16(OPT_TX_ENTRIES_PER_SUBCRQ);
5124 	ibmvnic_send_crq(adapter, &crq);
5125 	cap_reqs--;
5126 
5127 	crq.query_capability.capability =
5128 			cpu_to_be16(OPT_RXBA_ENTRIES_PER_SUBCRQ);
5129 	ibmvnic_send_crq(adapter, &crq);
5130 	cap_reqs--;
5131 
5132 	crq.query_capability.capability = cpu_to_be16(TX_RX_DESC_REQ);
5133 
5134 	ibmvnic_send_crq(adapter, &crq);
5135 	cap_reqs--;
5136 
5137 	/* Keep at end to catch any discrepancy between expected and actual
5138 	 * CRQs sent.
5139 	 */
5140 	WARN_ON(cap_reqs != 0);
5141 }
5142 
5143 static void send_query_ip_offload(struct ibmvnic_adapter *adapter)
5144 {
5145 	int buf_sz = sizeof(struct ibmvnic_query_ip_offload_buffer);
5146 	struct device *dev = &adapter->vdev->dev;
5147 	union ibmvnic_crq crq;
5148 
5149 	adapter->ip_offload_tok =
5150 		dma_map_single(dev,
5151 			       &adapter->ip_offload_buf,
5152 			       buf_sz,
5153 			       DMA_FROM_DEVICE);
5154 
5155 	if (dma_mapping_error(dev, adapter->ip_offload_tok)) {
5156 		if (!firmware_has_feature(FW_FEATURE_CMO))
5157 			dev_err(dev, "Couldn't map offload buffer\n");
5158 		return;
5159 	}
5160 
5161 	memset(&crq, 0, sizeof(crq));
5162 	crq.query_ip_offload.first = IBMVNIC_CRQ_CMD;
5163 	crq.query_ip_offload.cmd = QUERY_IP_OFFLOAD;
5164 	crq.query_ip_offload.len = cpu_to_be32(buf_sz);
5165 	crq.query_ip_offload.ioba =
5166 	    cpu_to_be32(adapter->ip_offload_tok);
5167 
5168 	ibmvnic_send_crq(adapter, &crq);
5169 }
5170 
5171 static void send_control_ip_offload(struct ibmvnic_adapter *adapter)
5172 {
5173 	struct ibmvnic_control_ip_offload_buffer *ctrl_buf = &adapter->ip_offload_ctrl;
5174 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5175 	struct device *dev = &adapter->vdev->dev;
5176 	netdev_features_t old_hw_features = 0;
5177 	union ibmvnic_crq crq;
5178 
5179 	adapter->ip_offload_ctrl_tok =
5180 		dma_map_single(dev,
5181 			       ctrl_buf,
5182 			       sizeof(adapter->ip_offload_ctrl),
5183 			       DMA_TO_DEVICE);
5184 
5185 	if (dma_mapping_error(dev, adapter->ip_offload_ctrl_tok)) {
5186 		dev_err(dev, "Couldn't map ip offload control buffer\n");
5187 		return;
5188 	}
5189 
5190 	ctrl_buf->len = cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5191 	ctrl_buf->version = cpu_to_be32(INITIAL_VERSION_IOB);
5192 	ctrl_buf->ipv4_chksum = buf->ipv4_chksum;
5193 	ctrl_buf->ipv6_chksum = buf->ipv6_chksum;
5194 	ctrl_buf->tcp_ipv4_chksum = buf->tcp_ipv4_chksum;
5195 	ctrl_buf->udp_ipv4_chksum = buf->udp_ipv4_chksum;
5196 	ctrl_buf->tcp_ipv6_chksum = buf->tcp_ipv6_chksum;
5197 	ctrl_buf->udp_ipv6_chksum = buf->udp_ipv6_chksum;
5198 	ctrl_buf->large_tx_ipv4 = buf->large_tx_ipv4;
5199 	ctrl_buf->large_tx_ipv6 = buf->large_tx_ipv6;
5200 
5201 	/* large_rx disabled for now, additional features needed */
5202 	ctrl_buf->large_rx_ipv4 = 0;
5203 	ctrl_buf->large_rx_ipv6 = 0;
5204 
5205 	if (adapter->state != VNIC_PROBING) {
5206 		old_hw_features = adapter->netdev->hw_features;
5207 		adapter->netdev->hw_features = 0;
5208 	}
5209 
5210 	adapter->netdev->hw_features = NETIF_F_SG | NETIF_F_GSO | NETIF_F_GRO;
5211 
5212 	if (buf->tcp_ipv4_chksum || buf->udp_ipv4_chksum)
5213 		adapter->netdev->hw_features |= NETIF_F_IP_CSUM;
5214 
5215 	if (buf->tcp_ipv6_chksum || buf->udp_ipv6_chksum)
5216 		adapter->netdev->hw_features |= NETIF_F_IPV6_CSUM;
5217 
5218 	if ((adapter->netdev->features &
5219 	    (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)))
5220 		adapter->netdev->hw_features |= NETIF_F_RXCSUM;
5221 
5222 	if (buf->large_tx_ipv4)
5223 		adapter->netdev->hw_features |= NETIF_F_TSO;
5224 	if (buf->large_tx_ipv6)
5225 		adapter->netdev->hw_features |= NETIF_F_TSO6;
5226 
5227 	if (adapter->state == VNIC_PROBING) {
5228 		adapter->netdev->features |= adapter->netdev->hw_features;
5229 	} else if (old_hw_features != adapter->netdev->hw_features) {
5230 		netdev_features_t tmp = 0;
5231 
5232 		/* disable features no longer supported */
5233 		adapter->netdev->features &= adapter->netdev->hw_features;
5234 		/* turn on features now supported if previously enabled */
5235 		tmp = (old_hw_features ^ adapter->netdev->hw_features) &
5236 			adapter->netdev->hw_features;
5237 		adapter->netdev->features |=
5238 				tmp & adapter->netdev->wanted_features;
5239 	}
5240 
5241 	memset(&crq, 0, sizeof(crq));
5242 	crq.control_ip_offload.first = IBMVNIC_CRQ_CMD;
5243 	crq.control_ip_offload.cmd = CONTROL_IP_OFFLOAD;
5244 	crq.control_ip_offload.len =
5245 	    cpu_to_be32(sizeof(adapter->ip_offload_ctrl));
5246 	crq.control_ip_offload.ioba = cpu_to_be32(adapter->ip_offload_ctrl_tok);
5247 	ibmvnic_send_crq(adapter, &crq);
5248 }
5249 
5250 static void handle_vpd_size_rsp(union ibmvnic_crq *crq,
5251 				struct ibmvnic_adapter *adapter)
5252 {
5253 	struct device *dev = &adapter->vdev->dev;
5254 
5255 	if (crq->get_vpd_size_rsp.rc.code) {
5256 		dev_err(dev, "Error retrieving VPD size, rc=%x\n",
5257 			crq->get_vpd_size_rsp.rc.code);
5258 		complete(&adapter->fw_done);
5259 		return;
5260 	}
5261 
5262 	adapter->vpd->len = be64_to_cpu(crq->get_vpd_size_rsp.len);
5263 	complete(&adapter->fw_done);
5264 }
5265 
5266 static void handle_vpd_rsp(union ibmvnic_crq *crq,
5267 			   struct ibmvnic_adapter *adapter)
5268 {
5269 	struct device *dev = &adapter->vdev->dev;
5270 	unsigned char *substr = NULL;
5271 	u8 fw_level_len = 0;
5272 
5273 	memset(adapter->fw_version, 0, 32);
5274 
5275 	dma_unmap_single(dev, adapter->vpd->dma_addr, adapter->vpd->len,
5276 			 DMA_FROM_DEVICE);
5277 
5278 	if (crq->get_vpd_rsp.rc.code) {
5279 		dev_err(dev, "Error retrieving VPD from device, rc=%x\n",
5280 			crq->get_vpd_rsp.rc.code);
5281 		goto complete;
5282 	}
5283 
5284 	/* get the position of the firmware version info
5285 	 * located after the ASCII 'RM' substring in the buffer
5286 	 */
5287 	substr = strnstr(adapter->vpd->buff, "RM", adapter->vpd->len);
5288 	if (!substr) {
5289 		dev_info(dev, "Warning - No FW level has been provided in the VPD buffer by the VIOS Server\n");
5290 		goto complete;
5291 	}
5292 
5293 	/* get length of firmware level ASCII substring */
5294 	if ((substr + 2) < (adapter->vpd->buff + adapter->vpd->len)) {
5295 		fw_level_len = *(substr + 2);
5296 	} else {
5297 		dev_info(dev, "Length of FW substr extrapolated VDP buff\n");
5298 		goto complete;
5299 	}
5300 
5301 	/* copy firmware version string from vpd into adapter */
5302 	if ((substr + 3 + fw_level_len) <
5303 	    (adapter->vpd->buff + adapter->vpd->len)) {
5304 		strscpy(adapter->fw_version, substr + 3,
5305 			sizeof(adapter->fw_version));
5306 	} else {
5307 		dev_info(dev, "FW substr extrapolated VPD buff\n");
5308 	}
5309 
5310 complete:
5311 	if (adapter->fw_version[0] == '\0')
5312 		strscpy((char *)adapter->fw_version, "N/A", sizeof(adapter->fw_version));
5313 	complete(&adapter->fw_done);
5314 }
5315 
5316 static void handle_query_ip_offload_rsp(struct ibmvnic_adapter *adapter)
5317 {
5318 	struct device *dev = &adapter->vdev->dev;
5319 	struct ibmvnic_query_ip_offload_buffer *buf = &adapter->ip_offload_buf;
5320 	int i;
5321 
5322 	dma_unmap_single(dev, adapter->ip_offload_tok,
5323 			 sizeof(adapter->ip_offload_buf), DMA_FROM_DEVICE);
5324 
5325 	netdev_dbg(adapter->netdev, "Query IP Offload Buffer:\n");
5326 	for (i = 0; i < (sizeof(adapter->ip_offload_buf) - 1) / 8 + 1; i++)
5327 		netdev_dbg(adapter->netdev, "%016lx\n",
5328 			   ((unsigned long *)(buf))[i]);
5329 
5330 	netdev_dbg(adapter->netdev, "ipv4_chksum = %d\n", buf->ipv4_chksum);
5331 	netdev_dbg(adapter->netdev, "ipv6_chksum = %d\n", buf->ipv6_chksum);
5332 	netdev_dbg(adapter->netdev, "tcp_ipv4_chksum = %d\n",
5333 		   buf->tcp_ipv4_chksum);
5334 	netdev_dbg(adapter->netdev, "tcp_ipv6_chksum = %d\n",
5335 		   buf->tcp_ipv6_chksum);
5336 	netdev_dbg(adapter->netdev, "udp_ipv4_chksum = %d\n",
5337 		   buf->udp_ipv4_chksum);
5338 	netdev_dbg(adapter->netdev, "udp_ipv6_chksum = %d\n",
5339 		   buf->udp_ipv6_chksum);
5340 	netdev_dbg(adapter->netdev, "large_tx_ipv4 = %d\n",
5341 		   buf->large_tx_ipv4);
5342 	netdev_dbg(adapter->netdev, "large_tx_ipv6 = %d\n",
5343 		   buf->large_tx_ipv6);
5344 	netdev_dbg(adapter->netdev, "large_rx_ipv4 = %d\n",
5345 		   buf->large_rx_ipv4);
5346 	netdev_dbg(adapter->netdev, "large_rx_ipv6 = %d\n",
5347 		   buf->large_rx_ipv6);
5348 	netdev_dbg(adapter->netdev, "max_ipv4_hdr_sz = %d\n",
5349 		   buf->max_ipv4_header_size);
5350 	netdev_dbg(adapter->netdev, "max_ipv6_hdr_sz = %d\n",
5351 		   buf->max_ipv6_header_size);
5352 	netdev_dbg(adapter->netdev, "max_tcp_hdr_size = %d\n",
5353 		   buf->max_tcp_header_size);
5354 	netdev_dbg(adapter->netdev, "max_udp_hdr_size = %d\n",
5355 		   buf->max_udp_header_size);
5356 	netdev_dbg(adapter->netdev, "max_large_tx_size = %d\n",
5357 		   buf->max_large_tx_size);
5358 	netdev_dbg(adapter->netdev, "max_large_rx_size = %d\n",
5359 		   buf->max_large_rx_size);
5360 	netdev_dbg(adapter->netdev, "ipv6_ext_hdr = %d\n",
5361 		   buf->ipv6_extension_header);
5362 	netdev_dbg(adapter->netdev, "tcp_pseudosum_req = %d\n",
5363 		   buf->tcp_pseudosum_req);
5364 	netdev_dbg(adapter->netdev, "num_ipv6_ext_hd = %d\n",
5365 		   buf->num_ipv6_ext_headers);
5366 	netdev_dbg(adapter->netdev, "off_ipv6_ext_hd = %d\n",
5367 		   buf->off_ipv6_ext_headers);
5368 
5369 	send_control_ip_offload(adapter);
5370 }
5371 
5372 static const char *ibmvnic_fw_err_cause(u16 cause)
5373 {
5374 	switch (cause) {
5375 	case ADAPTER_PROBLEM:
5376 		return "adapter problem";
5377 	case BUS_PROBLEM:
5378 		return "bus problem";
5379 	case FW_PROBLEM:
5380 		return "firmware problem";
5381 	case DD_PROBLEM:
5382 		return "device driver problem";
5383 	case EEH_RECOVERY:
5384 		return "EEH recovery";
5385 	case FW_UPDATED:
5386 		return "firmware updated";
5387 	case LOW_MEMORY:
5388 		return "low Memory";
5389 	default:
5390 		return "unknown";
5391 	}
5392 }
5393 
5394 static void handle_error_indication(union ibmvnic_crq *crq,
5395 				    struct ibmvnic_adapter *adapter)
5396 {
5397 	struct device *dev = &adapter->vdev->dev;
5398 	u16 cause;
5399 
5400 	cause = be16_to_cpu(crq->error_indication.error_cause);
5401 
5402 	dev_warn_ratelimited(dev,
5403 			     "Firmware reports %serror, cause: %s. Starting recovery...\n",
5404 			     crq->error_indication.flags
5405 				& IBMVNIC_FATAL_ERROR ? "FATAL " : "",
5406 			     ibmvnic_fw_err_cause(cause));
5407 
5408 	if (crq->error_indication.flags & IBMVNIC_FATAL_ERROR)
5409 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5410 	else
5411 		ibmvnic_reset(adapter, VNIC_RESET_NON_FATAL);
5412 }
5413 
5414 static int handle_change_mac_rsp(union ibmvnic_crq *crq,
5415 				 struct ibmvnic_adapter *adapter)
5416 {
5417 	struct net_device *netdev = adapter->netdev;
5418 	struct device *dev = &adapter->vdev->dev;
5419 	long rc;
5420 
5421 	rc = crq->change_mac_addr_rsp.rc.code;
5422 	if (rc) {
5423 		dev_err(dev, "Error %ld in CHANGE_MAC_ADDR_RSP\n", rc);
5424 		goto out;
5425 	}
5426 	/* crq->change_mac_addr.mac_addr is the requested one
5427 	 * crq->change_mac_addr_rsp.mac_addr is the returned valid one.
5428 	 */
5429 	eth_hw_addr_set(netdev, &crq->change_mac_addr_rsp.mac_addr[0]);
5430 	ether_addr_copy(adapter->mac_addr,
5431 			&crq->change_mac_addr_rsp.mac_addr[0]);
5432 out:
5433 	complete(&adapter->fw_done);
5434 	return rc;
5435 }
5436 
5437 static void handle_request_cap_rsp(union ibmvnic_crq *crq,
5438 				   struct ibmvnic_adapter *adapter)
5439 {
5440 	struct device *dev = &adapter->vdev->dev;
5441 	u64 *req_value;
5442 	char *name;
5443 
5444 	atomic_dec(&adapter->running_cap_crqs);
5445 	netdev_dbg(adapter->netdev, "Outstanding request-caps: %d\n",
5446 		   atomic_read(&adapter->running_cap_crqs));
5447 	switch (be16_to_cpu(crq->request_capability_rsp.capability)) {
5448 	case REQ_TX_QUEUES:
5449 		req_value = &adapter->req_tx_queues;
5450 		name = "tx";
5451 		break;
5452 	case REQ_RX_QUEUES:
5453 		req_value = &adapter->req_rx_queues;
5454 		name = "rx";
5455 		break;
5456 	case REQ_RX_ADD_QUEUES:
5457 		req_value = &adapter->req_rx_add_queues;
5458 		name = "rx_add";
5459 		break;
5460 	case REQ_TX_ENTRIES_PER_SUBCRQ:
5461 		req_value = &adapter->req_tx_entries_per_subcrq;
5462 		name = "tx_entries_per_subcrq";
5463 		break;
5464 	case REQ_RX_ADD_ENTRIES_PER_SUBCRQ:
5465 		req_value = &adapter->req_rx_add_entries_per_subcrq;
5466 		name = "rx_add_entries_per_subcrq";
5467 		break;
5468 	case REQ_MTU:
5469 		req_value = &adapter->req_mtu;
5470 		name = "mtu";
5471 		break;
5472 	case PROMISC_REQUESTED:
5473 		req_value = &adapter->promisc;
5474 		name = "promisc";
5475 		break;
5476 	default:
5477 		dev_err(dev, "Got invalid cap request rsp %d\n",
5478 			crq->request_capability.capability);
5479 		return;
5480 	}
5481 
5482 	switch (crq->request_capability_rsp.rc.code) {
5483 	case SUCCESS:
5484 		break;
5485 	case PARTIALSUCCESS:
5486 		dev_info(dev, "req=%lld, rsp=%ld in %s queue, retrying.\n",
5487 			 *req_value,
5488 			 (long)be64_to_cpu(crq->request_capability_rsp.number),
5489 			 name);
5490 
5491 		if (be16_to_cpu(crq->request_capability_rsp.capability) ==
5492 		    REQ_MTU) {
5493 			pr_err("mtu of %llu is not supported. Reverting.\n",
5494 			       *req_value);
5495 			*req_value = adapter->fallback.mtu;
5496 		} else {
5497 			*req_value =
5498 				be64_to_cpu(crq->request_capability_rsp.number);
5499 		}
5500 
5501 		send_request_cap(adapter, 1);
5502 		return;
5503 	default:
5504 		dev_err(dev, "Error %d in request cap rsp\n",
5505 			crq->request_capability_rsp.rc.code);
5506 		return;
5507 	}
5508 
5509 	/* Done receiving requested capabilities, query IP offload support */
5510 	if (atomic_read(&adapter->running_cap_crqs) == 0)
5511 		send_query_ip_offload(adapter);
5512 }
5513 
5514 static int handle_login_rsp(union ibmvnic_crq *login_rsp_crq,
5515 			    struct ibmvnic_adapter *adapter)
5516 {
5517 	struct device *dev = &adapter->vdev->dev;
5518 	struct net_device *netdev = adapter->netdev;
5519 	struct ibmvnic_login_rsp_buffer *login_rsp = adapter->login_rsp_buf;
5520 	struct ibmvnic_login_buffer *login = adapter->login_buf;
5521 	u64 *tx_handle_array;
5522 	u64 *rx_handle_array;
5523 	int num_tx_pools;
5524 	int num_rx_pools;
5525 	u64 *size_array;
5526 	u32 rsp_len;
5527 	int i;
5528 
5529 	/* CHECK: Test/set of login_pending does not need to be atomic
5530 	 * because only ibmvnic_tasklet tests/clears this.
5531 	 */
5532 	if (!adapter->login_pending) {
5533 		netdev_warn(netdev, "Ignoring unexpected login response\n");
5534 		return 0;
5535 	}
5536 	adapter->login_pending = false;
5537 
5538 	/* If the number of queues requested can't be allocated by the
5539 	 * server, the login response will return with code 1. We will need
5540 	 * to resend the login buffer with fewer queues requested.
5541 	 */
5542 	if (login_rsp_crq->generic.rc.code) {
5543 		adapter->init_done_rc = login_rsp_crq->generic.rc.code;
5544 		complete(&adapter->init_done);
5545 		return 0;
5546 	}
5547 
5548 	if (adapter->failover_pending) {
5549 		adapter->init_done_rc = -EAGAIN;
5550 		netdev_dbg(netdev, "Failover pending, ignoring login response\n");
5551 		complete(&adapter->init_done);
5552 		/* login response buffer will be released on reset */
5553 		return 0;
5554 	}
5555 
5556 	netdev->mtu = adapter->req_mtu - ETH_HLEN;
5557 
5558 	netdev_dbg(adapter->netdev, "Login Response Buffer:\n");
5559 	for (i = 0; i < (adapter->login_rsp_buf_sz - 1) / 8 + 1; i++) {
5560 		netdev_dbg(adapter->netdev, "%016lx\n",
5561 			   ((unsigned long *)(adapter->login_rsp_buf))[i]);
5562 	}
5563 
5564 	/* Sanity checks */
5565 	if (login->num_txcomp_subcrqs != login_rsp->num_txsubm_subcrqs ||
5566 	    (be32_to_cpu(login->num_rxcomp_subcrqs) *
5567 	     adapter->req_rx_add_queues !=
5568 	     be32_to_cpu(login_rsp->num_rxadd_subcrqs))) {
5569 		dev_err(dev, "FATAL: Inconsistent login and login rsp\n");
5570 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5571 		return -EIO;
5572 	}
5573 
5574 	rsp_len = be32_to_cpu(login_rsp->len);
5575 	if (be32_to_cpu(login->login_rsp_len) < rsp_len ||
5576 	    rsp_len <= be32_to_cpu(login_rsp->off_txsubm_subcrqs) ||
5577 	    rsp_len <= be32_to_cpu(login_rsp->off_rxadd_subcrqs) ||
5578 	    rsp_len <= be32_to_cpu(login_rsp->off_rxadd_buff_size) ||
5579 	    rsp_len <= be32_to_cpu(login_rsp->off_supp_tx_desc)) {
5580 		/* This can happen if a login request times out and there are
5581 		 * 2 outstanding login requests sent, the LOGIN_RSP crq
5582 		 * could have been for the older login request. So we are
5583 		 * parsing the newer response buffer which may be incomplete
5584 		 */
5585 		dev_err(dev, "FATAL: Login rsp offsets/lengths invalid\n");
5586 		ibmvnic_reset(adapter, VNIC_RESET_FATAL);
5587 		return -EIO;
5588 	}
5589 
5590 	size_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5591 		be32_to_cpu(adapter->login_rsp_buf->off_rxadd_buff_size));
5592 	/* variable buffer sizes are not supported, so just read the
5593 	 * first entry.
5594 	 */
5595 	adapter->cur_rx_buf_sz = be64_to_cpu(size_array[0]);
5596 
5597 	num_tx_pools = be32_to_cpu(adapter->login_rsp_buf->num_txsubm_subcrqs);
5598 	num_rx_pools = be32_to_cpu(adapter->login_rsp_buf->num_rxadd_subcrqs);
5599 
5600 	tx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5601 				  be32_to_cpu(adapter->login_rsp_buf->off_txsubm_subcrqs));
5602 	rx_handle_array = (u64 *)((u8 *)(adapter->login_rsp_buf) +
5603 				  be32_to_cpu(adapter->login_rsp_buf->off_rxadd_subcrqs));
5604 
5605 	for (i = 0; i < num_tx_pools; i++)
5606 		adapter->tx_scrq[i]->handle = tx_handle_array[i];
5607 
5608 	for (i = 0; i < num_rx_pools; i++)
5609 		adapter->rx_scrq[i]->handle = rx_handle_array[i];
5610 
5611 	adapter->num_active_tx_scrqs = num_tx_pools;
5612 	adapter->num_active_rx_scrqs = num_rx_pools;
5613 	release_login_rsp_buffer(adapter);
5614 	release_login_buffer(adapter);
5615 	complete(&adapter->init_done);
5616 
5617 	return 0;
5618 }
5619 
5620 static void handle_request_unmap_rsp(union ibmvnic_crq *crq,
5621 				     struct ibmvnic_adapter *adapter)
5622 {
5623 	struct device *dev = &adapter->vdev->dev;
5624 	long rc;
5625 
5626 	rc = crq->request_unmap_rsp.rc.code;
5627 	if (rc)
5628 		dev_err(dev, "Error %ld in REQUEST_UNMAP_RSP\n", rc);
5629 }
5630 
5631 static void handle_query_map_rsp(union ibmvnic_crq *crq,
5632 				 struct ibmvnic_adapter *adapter)
5633 {
5634 	struct net_device *netdev = adapter->netdev;
5635 	struct device *dev = &adapter->vdev->dev;
5636 	long rc;
5637 
5638 	rc = crq->query_map_rsp.rc.code;
5639 	if (rc) {
5640 		dev_err(dev, "Error %ld in QUERY_MAP_RSP\n", rc);
5641 		return;
5642 	}
5643 	netdev_dbg(netdev, "page_size = %d\ntot_pages = %u\nfree_pages = %u\n",
5644 		   crq->query_map_rsp.page_size,
5645 		   __be32_to_cpu(crq->query_map_rsp.tot_pages),
5646 		   __be32_to_cpu(crq->query_map_rsp.free_pages));
5647 }
5648 
5649 static void handle_query_cap_rsp(union ibmvnic_crq *crq,
5650 				 struct ibmvnic_adapter *adapter)
5651 {
5652 	struct net_device *netdev = adapter->netdev;
5653 	struct device *dev = &adapter->vdev->dev;
5654 	long rc;
5655 
5656 	atomic_dec(&adapter->running_cap_crqs);
5657 	netdev_dbg(netdev, "Outstanding queries: %d\n",
5658 		   atomic_read(&adapter->running_cap_crqs));
5659 	rc = crq->query_capability.rc.code;
5660 	if (rc) {
5661 		dev_err(dev, "Error %ld in QUERY_CAP_RSP\n", rc);
5662 		goto out;
5663 	}
5664 
5665 	switch (be16_to_cpu(crq->query_capability.capability)) {
5666 	case MIN_TX_QUEUES:
5667 		adapter->min_tx_queues =
5668 		    be64_to_cpu(crq->query_capability.number);
5669 		netdev_dbg(netdev, "min_tx_queues = %lld\n",
5670 			   adapter->min_tx_queues);
5671 		break;
5672 	case MIN_RX_QUEUES:
5673 		adapter->min_rx_queues =
5674 		    be64_to_cpu(crq->query_capability.number);
5675 		netdev_dbg(netdev, "min_rx_queues = %lld\n",
5676 			   adapter->min_rx_queues);
5677 		break;
5678 	case MIN_RX_ADD_QUEUES:
5679 		adapter->min_rx_add_queues =
5680 		    be64_to_cpu(crq->query_capability.number);
5681 		netdev_dbg(netdev, "min_rx_add_queues = %lld\n",
5682 			   adapter->min_rx_add_queues);
5683 		break;
5684 	case MAX_TX_QUEUES:
5685 		adapter->max_tx_queues =
5686 		    be64_to_cpu(crq->query_capability.number);
5687 		netdev_dbg(netdev, "max_tx_queues = %lld\n",
5688 			   adapter->max_tx_queues);
5689 		break;
5690 	case MAX_RX_QUEUES:
5691 		adapter->max_rx_queues =
5692 		    be64_to_cpu(crq->query_capability.number);
5693 		netdev_dbg(netdev, "max_rx_queues = %lld\n",
5694 			   adapter->max_rx_queues);
5695 		break;
5696 	case MAX_RX_ADD_QUEUES:
5697 		adapter->max_rx_add_queues =
5698 		    be64_to_cpu(crq->query_capability.number);
5699 		netdev_dbg(netdev, "max_rx_add_queues = %lld\n",
5700 			   adapter->max_rx_add_queues);
5701 		break;
5702 	case MIN_TX_ENTRIES_PER_SUBCRQ:
5703 		adapter->min_tx_entries_per_subcrq =
5704 		    be64_to_cpu(crq->query_capability.number);
5705 		netdev_dbg(netdev, "min_tx_entries_per_subcrq = %lld\n",
5706 			   adapter->min_tx_entries_per_subcrq);
5707 		break;
5708 	case MIN_RX_ADD_ENTRIES_PER_SUBCRQ:
5709 		adapter->min_rx_add_entries_per_subcrq =
5710 		    be64_to_cpu(crq->query_capability.number);
5711 		netdev_dbg(netdev, "min_rx_add_entrs_per_subcrq = %lld\n",
5712 			   adapter->min_rx_add_entries_per_subcrq);
5713 		break;
5714 	case MAX_TX_ENTRIES_PER_SUBCRQ:
5715 		adapter->max_tx_entries_per_subcrq =
5716 		    be64_to_cpu(crq->query_capability.number);
5717 		netdev_dbg(netdev, "max_tx_entries_per_subcrq = %lld\n",
5718 			   adapter->max_tx_entries_per_subcrq);
5719 		break;
5720 	case MAX_RX_ADD_ENTRIES_PER_SUBCRQ:
5721 		adapter->max_rx_add_entries_per_subcrq =
5722 		    be64_to_cpu(crq->query_capability.number);
5723 		netdev_dbg(netdev, "max_rx_add_entrs_per_subcrq = %lld\n",
5724 			   adapter->max_rx_add_entries_per_subcrq);
5725 		break;
5726 	case TCP_IP_OFFLOAD:
5727 		adapter->tcp_ip_offload =
5728 		    be64_to_cpu(crq->query_capability.number);
5729 		netdev_dbg(netdev, "tcp_ip_offload = %lld\n",
5730 			   adapter->tcp_ip_offload);
5731 		break;
5732 	case PROMISC_SUPPORTED:
5733 		adapter->promisc_supported =
5734 		    be64_to_cpu(crq->query_capability.number);
5735 		netdev_dbg(netdev, "promisc_supported = %lld\n",
5736 			   adapter->promisc_supported);
5737 		break;
5738 	case MIN_MTU:
5739 		adapter->min_mtu = be64_to_cpu(crq->query_capability.number);
5740 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
5741 		netdev_dbg(netdev, "min_mtu = %lld\n", adapter->min_mtu);
5742 		break;
5743 	case MAX_MTU:
5744 		adapter->max_mtu = be64_to_cpu(crq->query_capability.number);
5745 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
5746 		netdev_dbg(netdev, "max_mtu = %lld\n", adapter->max_mtu);
5747 		break;
5748 	case MAX_MULTICAST_FILTERS:
5749 		adapter->max_multicast_filters =
5750 		    be64_to_cpu(crq->query_capability.number);
5751 		netdev_dbg(netdev, "max_multicast_filters = %lld\n",
5752 			   adapter->max_multicast_filters);
5753 		break;
5754 	case VLAN_HEADER_INSERTION:
5755 		adapter->vlan_header_insertion =
5756 		    be64_to_cpu(crq->query_capability.number);
5757 		if (adapter->vlan_header_insertion)
5758 			netdev->features |= NETIF_F_HW_VLAN_STAG_TX;
5759 		netdev_dbg(netdev, "vlan_header_insertion = %lld\n",
5760 			   adapter->vlan_header_insertion);
5761 		break;
5762 	case RX_VLAN_HEADER_INSERTION:
5763 		adapter->rx_vlan_header_insertion =
5764 		    be64_to_cpu(crq->query_capability.number);
5765 		netdev_dbg(netdev, "rx_vlan_header_insertion = %lld\n",
5766 			   adapter->rx_vlan_header_insertion);
5767 		break;
5768 	case MAX_TX_SG_ENTRIES:
5769 		adapter->max_tx_sg_entries =
5770 		    be64_to_cpu(crq->query_capability.number);
5771 		netdev_dbg(netdev, "max_tx_sg_entries = %lld\n",
5772 			   adapter->max_tx_sg_entries);
5773 		break;
5774 	case RX_SG_SUPPORTED:
5775 		adapter->rx_sg_supported =
5776 		    be64_to_cpu(crq->query_capability.number);
5777 		netdev_dbg(netdev, "rx_sg_supported = %lld\n",
5778 			   adapter->rx_sg_supported);
5779 		break;
5780 	case OPT_TX_COMP_SUB_QUEUES:
5781 		adapter->opt_tx_comp_sub_queues =
5782 		    be64_to_cpu(crq->query_capability.number);
5783 		netdev_dbg(netdev, "opt_tx_comp_sub_queues = %lld\n",
5784 			   adapter->opt_tx_comp_sub_queues);
5785 		break;
5786 	case OPT_RX_COMP_QUEUES:
5787 		adapter->opt_rx_comp_queues =
5788 		    be64_to_cpu(crq->query_capability.number);
5789 		netdev_dbg(netdev, "opt_rx_comp_queues = %lld\n",
5790 			   adapter->opt_rx_comp_queues);
5791 		break;
5792 	case OPT_RX_BUFADD_Q_PER_RX_COMP_Q:
5793 		adapter->opt_rx_bufadd_q_per_rx_comp_q =
5794 		    be64_to_cpu(crq->query_capability.number);
5795 		netdev_dbg(netdev, "opt_rx_bufadd_q_per_rx_comp_q = %lld\n",
5796 			   adapter->opt_rx_bufadd_q_per_rx_comp_q);
5797 		break;
5798 	case OPT_TX_ENTRIES_PER_SUBCRQ:
5799 		adapter->opt_tx_entries_per_subcrq =
5800 		    be64_to_cpu(crq->query_capability.number);
5801 		netdev_dbg(netdev, "opt_tx_entries_per_subcrq = %lld\n",
5802 			   adapter->opt_tx_entries_per_subcrq);
5803 		break;
5804 	case OPT_RXBA_ENTRIES_PER_SUBCRQ:
5805 		adapter->opt_rxba_entries_per_subcrq =
5806 		    be64_to_cpu(crq->query_capability.number);
5807 		netdev_dbg(netdev, "opt_rxba_entries_per_subcrq = %lld\n",
5808 			   adapter->opt_rxba_entries_per_subcrq);
5809 		break;
5810 	case TX_RX_DESC_REQ:
5811 		adapter->tx_rx_desc_req = crq->query_capability.number;
5812 		netdev_dbg(netdev, "tx_rx_desc_req = %llx\n",
5813 			   adapter->tx_rx_desc_req);
5814 		break;
5815 
5816 	default:
5817 		netdev_err(netdev, "Got invalid cap rsp %d\n",
5818 			   crq->query_capability.capability);
5819 	}
5820 
5821 out:
5822 	if (atomic_read(&adapter->running_cap_crqs) == 0)
5823 		send_request_cap(adapter, 0);
5824 }
5825 
5826 static int send_query_phys_parms(struct ibmvnic_adapter *adapter)
5827 {
5828 	union ibmvnic_crq crq;
5829 	int rc;
5830 
5831 	memset(&crq, 0, sizeof(crq));
5832 	crq.query_phys_parms.first = IBMVNIC_CRQ_CMD;
5833 	crq.query_phys_parms.cmd = QUERY_PHYS_PARMS;
5834 
5835 	mutex_lock(&adapter->fw_lock);
5836 	adapter->fw_done_rc = 0;
5837 	reinit_completion(&adapter->fw_done);
5838 
5839 	rc = ibmvnic_send_crq(adapter, &crq);
5840 	if (rc) {
5841 		mutex_unlock(&adapter->fw_lock);
5842 		return rc;
5843 	}
5844 
5845 	rc = ibmvnic_wait_for_completion(adapter, &adapter->fw_done, 10000);
5846 	if (rc) {
5847 		mutex_unlock(&adapter->fw_lock);
5848 		return rc;
5849 	}
5850 
5851 	mutex_unlock(&adapter->fw_lock);
5852 	return adapter->fw_done_rc ? -EIO : 0;
5853 }
5854 
5855 static int handle_query_phys_parms_rsp(union ibmvnic_crq *crq,
5856 				       struct ibmvnic_adapter *adapter)
5857 {
5858 	struct net_device *netdev = adapter->netdev;
5859 	int rc;
5860 	__be32 rspeed = cpu_to_be32(crq->query_phys_parms_rsp.speed);
5861 
5862 	rc = crq->query_phys_parms_rsp.rc.code;
5863 	if (rc) {
5864 		netdev_err(netdev, "Error %d in QUERY_PHYS_PARMS\n", rc);
5865 		return rc;
5866 	}
5867 	switch (rspeed) {
5868 	case IBMVNIC_10MBPS:
5869 		adapter->speed = SPEED_10;
5870 		break;
5871 	case IBMVNIC_100MBPS:
5872 		adapter->speed = SPEED_100;
5873 		break;
5874 	case IBMVNIC_1GBPS:
5875 		adapter->speed = SPEED_1000;
5876 		break;
5877 	case IBMVNIC_10GBPS:
5878 		adapter->speed = SPEED_10000;
5879 		break;
5880 	case IBMVNIC_25GBPS:
5881 		adapter->speed = SPEED_25000;
5882 		break;
5883 	case IBMVNIC_40GBPS:
5884 		adapter->speed = SPEED_40000;
5885 		break;
5886 	case IBMVNIC_50GBPS:
5887 		adapter->speed = SPEED_50000;
5888 		break;
5889 	case IBMVNIC_100GBPS:
5890 		adapter->speed = SPEED_100000;
5891 		break;
5892 	case IBMVNIC_200GBPS:
5893 		adapter->speed = SPEED_200000;
5894 		break;
5895 	default:
5896 		if (netif_carrier_ok(netdev))
5897 			netdev_warn(netdev, "Unknown speed 0x%08x\n", rspeed);
5898 		adapter->speed = SPEED_UNKNOWN;
5899 	}
5900 	if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_FULL_DUPLEX)
5901 		adapter->duplex = DUPLEX_FULL;
5902 	else if (crq->query_phys_parms_rsp.flags1 & IBMVNIC_HALF_DUPLEX)
5903 		adapter->duplex = DUPLEX_HALF;
5904 	else
5905 		adapter->duplex = DUPLEX_UNKNOWN;
5906 
5907 	return rc;
5908 }
5909 
5910 static void ibmvnic_handle_crq(union ibmvnic_crq *crq,
5911 			       struct ibmvnic_adapter *adapter)
5912 {
5913 	struct ibmvnic_generic_crq *gen_crq = &crq->generic;
5914 	struct net_device *netdev = adapter->netdev;
5915 	struct device *dev = &adapter->vdev->dev;
5916 	u64 *u64_crq = (u64 *)crq;
5917 	long rc;
5918 
5919 	netdev_dbg(netdev, "Handling CRQ: %016lx %016lx\n",
5920 		   (unsigned long)cpu_to_be64(u64_crq[0]),
5921 		   (unsigned long)cpu_to_be64(u64_crq[1]));
5922 	switch (gen_crq->first) {
5923 	case IBMVNIC_CRQ_INIT_RSP:
5924 		switch (gen_crq->cmd) {
5925 		case IBMVNIC_CRQ_INIT:
5926 			dev_info(dev, "Partner initialized\n");
5927 			adapter->from_passive_init = true;
5928 			/* Discard any stale login responses from prev reset.
5929 			 * CHECK: should we clear even on INIT_COMPLETE?
5930 			 */
5931 			adapter->login_pending = false;
5932 
5933 			if (adapter->state == VNIC_DOWN)
5934 				rc = ibmvnic_reset(adapter, VNIC_RESET_PASSIVE_INIT);
5935 			else
5936 				rc = ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
5937 
5938 			if (rc && rc != -EBUSY) {
5939 				/* We were unable to schedule the failover
5940 				 * reset either because the adapter was still
5941 				 * probing (eg: during kexec) or we could not
5942 				 * allocate memory. Clear the failover_pending
5943 				 * flag since no one else will. We ignore
5944 				 * EBUSY because it means either FAILOVER reset
5945 				 * is already scheduled or the adapter is
5946 				 * being removed.
5947 				 */
5948 				netdev_err(netdev,
5949 					   "Error %ld scheduling failover reset\n",
5950 					   rc);
5951 				adapter->failover_pending = false;
5952 			}
5953 
5954 			if (!completion_done(&adapter->init_done)) {
5955 				if (!adapter->init_done_rc)
5956 					adapter->init_done_rc = -EAGAIN;
5957 				complete(&adapter->init_done);
5958 			}
5959 
5960 			break;
5961 		case IBMVNIC_CRQ_INIT_COMPLETE:
5962 			dev_info(dev, "Partner initialization complete\n");
5963 			adapter->crq.active = true;
5964 			send_version_xchg(adapter);
5965 			break;
5966 		default:
5967 			dev_err(dev, "Unknown crq cmd: %d\n", gen_crq->cmd);
5968 		}
5969 		return;
5970 	case IBMVNIC_CRQ_XPORT_EVENT:
5971 		netif_carrier_off(netdev);
5972 		adapter->crq.active = false;
5973 		/* terminate any thread waiting for a response
5974 		 * from the device
5975 		 */
5976 		if (!completion_done(&adapter->fw_done)) {
5977 			adapter->fw_done_rc = -EIO;
5978 			complete(&adapter->fw_done);
5979 		}
5980 
5981 		/* if we got here during crq-init, retry crq-init */
5982 		if (!completion_done(&adapter->init_done)) {
5983 			adapter->init_done_rc = -EAGAIN;
5984 			complete(&adapter->init_done);
5985 		}
5986 
5987 		if (!completion_done(&adapter->stats_done))
5988 			complete(&adapter->stats_done);
5989 		if (test_bit(0, &adapter->resetting))
5990 			adapter->force_reset_recovery = true;
5991 		if (gen_crq->cmd == IBMVNIC_PARTITION_MIGRATED) {
5992 			dev_info(dev, "Migrated, re-enabling adapter\n");
5993 			ibmvnic_reset(adapter, VNIC_RESET_MOBILITY);
5994 		} else if (gen_crq->cmd == IBMVNIC_DEVICE_FAILOVER) {
5995 			dev_info(dev, "Backing device failover detected\n");
5996 			adapter->failover_pending = true;
5997 		} else {
5998 			/* The adapter lost the connection */
5999 			dev_err(dev, "Virtual Adapter failed (rc=%d)\n",
6000 				gen_crq->cmd);
6001 			ibmvnic_reset(adapter, VNIC_RESET_FATAL);
6002 		}
6003 		return;
6004 	case IBMVNIC_CRQ_CMD_RSP:
6005 		break;
6006 	default:
6007 		dev_err(dev, "Got an invalid msg type 0x%02x\n",
6008 			gen_crq->first);
6009 		return;
6010 	}
6011 
6012 	switch (gen_crq->cmd) {
6013 	case VERSION_EXCHANGE_RSP:
6014 		rc = crq->version_exchange_rsp.rc.code;
6015 		if (rc) {
6016 			dev_err(dev, "Error %ld in VERSION_EXCHG_RSP\n", rc);
6017 			break;
6018 		}
6019 		ibmvnic_version =
6020 			    be16_to_cpu(crq->version_exchange_rsp.version);
6021 		dev_info(dev, "Partner protocol version is %d\n",
6022 			 ibmvnic_version);
6023 		send_query_cap(adapter);
6024 		break;
6025 	case QUERY_CAPABILITY_RSP:
6026 		handle_query_cap_rsp(crq, adapter);
6027 		break;
6028 	case QUERY_MAP_RSP:
6029 		handle_query_map_rsp(crq, adapter);
6030 		break;
6031 	case REQUEST_MAP_RSP:
6032 		adapter->fw_done_rc = crq->request_map_rsp.rc.code;
6033 		complete(&adapter->fw_done);
6034 		break;
6035 	case REQUEST_UNMAP_RSP:
6036 		handle_request_unmap_rsp(crq, adapter);
6037 		break;
6038 	case REQUEST_CAPABILITY_RSP:
6039 		handle_request_cap_rsp(crq, adapter);
6040 		break;
6041 	case LOGIN_RSP:
6042 		netdev_dbg(netdev, "Got Login Response\n");
6043 		handle_login_rsp(crq, adapter);
6044 		break;
6045 	case LOGICAL_LINK_STATE_RSP:
6046 		netdev_dbg(netdev,
6047 			   "Got Logical Link State Response, state: %d rc: %d\n",
6048 			   crq->logical_link_state_rsp.link_state,
6049 			   crq->logical_link_state_rsp.rc.code);
6050 		adapter->logical_link_state =
6051 		    crq->logical_link_state_rsp.link_state;
6052 		adapter->init_done_rc = crq->logical_link_state_rsp.rc.code;
6053 		complete(&adapter->init_done);
6054 		break;
6055 	case LINK_STATE_INDICATION:
6056 		netdev_dbg(netdev, "Got Logical Link State Indication\n");
6057 		adapter->phys_link_state =
6058 		    crq->link_state_indication.phys_link_state;
6059 		adapter->logical_link_state =
6060 		    crq->link_state_indication.logical_link_state;
6061 		if (adapter->phys_link_state && adapter->logical_link_state)
6062 			netif_carrier_on(netdev);
6063 		else
6064 			netif_carrier_off(netdev);
6065 		break;
6066 	case CHANGE_MAC_ADDR_RSP:
6067 		netdev_dbg(netdev, "Got MAC address change Response\n");
6068 		adapter->fw_done_rc = handle_change_mac_rsp(crq, adapter);
6069 		break;
6070 	case ERROR_INDICATION:
6071 		netdev_dbg(netdev, "Got Error Indication\n");
6072 		handle_error_indication(crq, adapter);
6073 		break;
6074 	case REQUEST_STATISTICS_RSP:
6075 		netdev_dbg(netdev, "Got Statistics Response\n");
6076 		complete(&adapter->stats_done);
6077 		break;
6078 	case QUERY_IP_OFFLOAD_RSP:
6079 		netdev_dbg(netdev, "Got Query IP offload Response\n");
6080 		handle_query_ip_offload_rsp(adapter);
6081 		break;
6082 	case MULTICAST_CTRL_RSP:
6083 		netdev_dbg(netdev, "Got multicast control Response\n");
6084 		break;
6085 	case CONTROL_IP_OFFLOAD_RSP:
6086 		netdev_dbg(netdev, "Got Control IP offload Response\n");
6087 		dma_unmap_single(dev, adapter->ip_offload_ctrl_tok,
6088 				 sizeof(adapter->ip_offload_ctrl),
6089 				 DMA_TO_DEVICE);
6090 		complete(&adapter->init_done);
6091 		break;
6092 	case COLLECT_FW_TRACE_RSP:
6093 		netdev_dbg(netdev, "Got Collect firmware trace Response\n");
6094 		complete(&adapter->fw_done);
6095 		break;
6096 	case GET_VPD_SIZE_RSP:
6097 		handle_vpd_size_rsp(crq, adapter);
6098 		break;
6099 	case GET_VPD_RSP:
6100 		handle_vpd_rsp(crq, adapter);
6101 		break;
6102 	case QUERY_PHYS_PARMS_RSP:
6103 		adapter->fw_done_rc = handle_query_phys_parms_rsp(crq, adapter);
6104 		complete(&adapter->fw_done);
6105 		break;
6106 	default:
6107 		netdev_err(netdev, "Got an invalid cmd type 0x%02x\n",
6108 			   gen_crq->cmd);
6109 	}
6110 }
6111 
6112 static irqreturn_t ibmvnic_interrupt(int irq, void *instance)
6113 {
6114 	struct ibmvnic_adapter *adapter = instance;
6115 
6116 	tasklet_schedule(&adapter->tasklet);
6117 	return IRQ_HANDLED;
6118 }
6119 
6120 static void ibmvnic_tasklet(struct tasklet_struct *t)
6121 {
6122 	struct ibmvnic_adapter *adapter = from_tasklet(adapter, t, tasklet);
6123 	struct ibmvnic_crq_queue *queue = &adapter->crq;
6124 	union ibmvnic_crq *crq;
6125 	unsigned long flags;
6126 
6127 	spin_lock_irqsave(&queue->lock, flags);
6128 
6129 	/* Pull all the valid messages off the CRQ */
6130 	while ((crq = ibmvnic_next_crq(adapter)) != NULL) {
6131 		/* This barrier makes sure ibmvnic_next_crq()'s
6132 		 * crq->generic.first & IBMVNIC_CRQ_CMD_RSP is loaded
6133 		 * before ibmvnic_handle_crq()'s
6134 		 * switch(gen_crq->first) and switch(gen_crq->cmd).
6135 		 */
6136 		dma_rmb();
6137 		ibmvnic_handle_crq(crq, adapter);
6138 		crq->generic.first = 0;
6139 	}
6140 
6141 	spin_unlock_irqrestore(&queue->lock, flags);
6142 }
6143 
6144 static int ibmvnic_reenable_crq_queue(struct ibmvnic_adapter *adapter)
6145 {
6146 	struct vio_dev *vdev = adapter->vdev;
6147 	int rc;
6148 
6149 	do {
6150 		rc = plpar_hcall_norets(H_ENABLE_CRQ, vdev->unit_address);
6151 	} while (rc == H_IN_PROGRESS || rc == H_BUSY || H_IS_LONG_BUSY(rc));
6152 
6153 	if (rc)
6154 		dev_err(&vdev->dev, "Error enabling adapter (rc=%d)\n", rc);
6155 
6156 	return rc;
6157 }
6158 
6159 static int ibmvnic_reset_crq(struct ibmvnic_adapter *adapter)
6160 {
6161 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6162 	struct device *dev = &adapter->vdev->dev;
6163 	struct vio_dev *vdev = adapter->vdev;
6164 	int rc;
6165 
6166 	/* Close the CRQ */
6167 	do {
6168 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6169 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6170 
6171 	/* Clean out the queue */
6172 	if (!crq->msgs)
6173 		return -EINVAL;
6174 
6175 	memset(crq->msgs, 0, PAGE_SIZE);
6176 	crq->cur = 0;
6177 	crq->active = false;
6178 
6179 	/* And re-open it again */
6180 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6181 				crq->msg_token, PAGE_SIZE);
6182 
6183 	if (rc == H_CLOSED)
6184 		/* Adapter is good, but other end is not ready */
6185 		dev_warn(dev, "Partner adapter not ready\n");
6186 	else if (rc != 0)
6187 		dev_warn(dev, "Couldn't register crq (rc=%d)\n", rc);
6188 
6189 	return rc;
6190 }
6191 
6192 static void release_crq_queue(struct ibmvnic_adapter *adapter)
6193 {
6194 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6195 	struct vio_dev *vdev = adapter->vdev;
6196 	long rc;
6197 
6198 	if (!crq->msgs)
6199 		return;
6200 
6201 	netdev_dbg(adapter->netdev, "Releasing CRQ\n");
6202 	free_irq(vdev->irq, adapter);
6203 	tasklet_kill(&adapter->tasklet);
6204 	do {
6205 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6206 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6207 
6208 	dma_unmap_single(&vdev->dev, crq->msg_token, PAGE_SIZE,
6209 			 DMA_BIDIRECTIONAL);
6210 	free_page((unsigned long)crq->msgs);
6211 	crq->msgs = NULL;
6212 	crq->active = false;
6213 }
6214 
6215 static int init_crq_queue(struct ibmvnic_adapter *adapter)
6216 {
6217 	struct ibmvnic_crq_queue *crq = &adapter->crq;
6218 	struct device *dev = &adapter->vdev->dev;
6219 	struct vio_dev *vdev = adapter->vdev;
6220 	int rc, retrc = -ENOMEM;
6221 
6222 	if (crq->msgs)
6223 		return 0;
6224 
6225 	crq->msgs = (union ibmvnic_crq *)get_zeroed_page(GFP_KERNEL);
6226 	/* Should we allocate more than one page? */
6227 
6228 	if (!crq->msgs)
6229 		return -ENOMEM;
6230 
6231 	crq->size = PAGE_SIZE / sizeof(*crq->msgs);
6232 	crq->msg_token = dma_map_single(dev, crq->msgs, PAGE_SIZE,
6233 					DMA_BIDIRECTIONAL);
6234 	if (dma_mapping_error(dev, crq->msg_token))
6235 		goto map_failed;
6236 
6237 	rc = plpar_hcall_norets(H_REG_CRQ, vdev->unit_address,
6238 				crq->msg_token, PAGE_SIZE);
6239 
6240 	if (rc == H_RESOURCE)
6241 		/* maybe kexecing and resource is busy. try a reset */
6242 		rc = ibmvnic_reset_crq(adapter);
6243 	retrc = rc;
6244 
6245 	if (rc == H_CLOSED) {
6246 		dev_warn(dev, "Partner adapter not ready\n");
6247 	} else if (rc) {
6248 		dev_warn(dev, "Error %d opening adapter\n", rc);
6249 		goto reg_crq_failed;
6250 	}
6251 
6252 	retrc = 0;
6253 
6254 	tasklet_setup(&adapter->tasklet, (void *)ibmvnic_tasklet);
6255 
6256 	netdev_dbg(adapter->netdev, "registering irq 0x%x\n", vdev->irq);
6257 	snprintf(crq->name, sizeof(crq->name), "ibmvnic-%x",
6258 		 adapter->vdev->unit_address);
6259 	rc = request_irq(vdev->irq, ibmvnic_interrupt, 0, crq->name, adapter);
6260 	if (rc) {
6261 		dev_err(dev, "Couldn't register irq 0x%x. rc=%d\n",
6262 			vdev->irq, rc);
6263 		goto req_irq_failed;
6264 	}
6265 
6266 	rc = vio_enable_interrupts(vdev);
6267 	if (rc) {
6268 		dev_err(dev, "Error %d enabling interrupts\n", rc);
6269 		goto req_irq_failed;
6270 	}
6271 
6272 	crq->cur = 0;
6273 	spin_lock_init(&crq->lock);
6274 
6275 	/* process any CRQs that were queued before we enabled interrupts */
6276 	tasklet_schedule(&adapter->tasklet);
6277 
6278 	return retrc;
6279 
6280 req_irq_failed:
6281 	tasklet_kill(&adapter->tasklet);
6282 	do {
6283 		rc = plpar_hcall_norets(H_FREE_CRQ, vdev->unit_address);
6284 	} while (rc == H_BUSY || H_IS_LONG_BUSY(rc));
6285 reg_crq_failed:
6286 	dma_unmap_single(dev, crq->msg_token, PAGE_SIZE, DMA_BIDIRECTIONAL);
6287 map_failed:
6288 	free_page((unsigned long)crq->msgs);
6289 	crq->msgs = NULL;
6290 	return retrc;
6291 }
6292 
6293 static int ibmvnic_reset_init(struct ibmvnic_adapter *adapter, bool reset)
6294 {
6295 	struct device *dev = &adapter->vdev->dev;
6296 	unsigned long timeout = msecs_to_jiffies(20000);
6297 	u64 old_num_rx_queues = adapter->req_rx_queues;
6298 	u64 old_num_tx_queues = adapter->req_tx_queues;
6299 	int rc;
6300 
6301 	adapter->from_passive_init = false;
6302 
6303 	rc = ibmvnic_send_crq_init(adapter);
6304 	if (rc) {
6305 		dev_err(dev, "Send crq init failed with error %d\n", rc);
6306 		return rc;
6307 	}
6308 
6309 	if (!wait_for_completion_timeout(&adapter->init_done, timeout)) {
6310 		dev_err(dev, "Initialization sequence timed out\n");
6311 		return -ETIMEDOUT;
6312 	}
6313 
6314 	if (adapter->init_done_rc) {
6315 		release_crq_queue(adapter);
6316 		dev_err(dev, "CRQ-init failed, %d\n", adapter->init_done_rc);
6317 		return adapter->init_done_rc;
6318 	}
6319 
6320 	if (adapter->from_passive_init) {
6321 		adapter->state = VNIC_OPEN;
6322 		adapter->from_passive_init = false;
6323 		dev_err(dev, "CRQ-init failed, passive-init\n");
6324 		return -EINVAL;
6325 	}
6326 
6327 	if (reset &&
6328 	    test_bit(0, &adapter->resetting) && !adapter->wait_for_reset &&
6329 	    adapter->reset_reason != VNIC_RESET_MOBILITY) {
6330 		if (adapter->req_rx_queues != old_num_rx_queues ||
6331 		    adapter->req_tx_queues != old_num_tx_queues) {
6332 			release_sub_crqs(adapter, 0);
6333 			rc = init_sub_crqs(adapter);
6334 		} else {
6335 			/* no need to reinitialize completely, but we do
6336 			 * need to clean up transmits that were in flight
6337 			 * when we processed the reset.  Failure to do so
6338 			 * will confound the upper layer, usually TCP, by
6339 			 * creating the illusion of transmits that are
6340 			 * awaiting completion.
6341 			 */
6342 			clean_tx_pools(adapter);
6343 
6344 			rc = reset_sub_crq_queues(adapter);
6345 		}
6346 	} else {
6347 		rc = init_sub_crqs(adapter);
6348 	}
6349 
6350 	if (rc) {
6351 		dev_err(dev, "Initialization of sub crqs failed\n");
6352 		release_crq_queue(adapter);
6353 		return rc;
6354 	}
6355 
6356 	rc = init_sub_crq_irqs(adapter);
6357 	if (rc) {
6358 		dev_err(dev, "Failed to initialize sub crq irqs\n");
6359 		release_crq_queue(adapter);
6360 	}
6361 
6362 	return rc;
6363 }
6364 
6365 static struct device_attribute dev_attr_failover;
6366 
6367 static int ibmvnic_probe(struct vio_dev *dev, const struct vio_device_id *id)
6368 {
6369 	struct ibmvnic_adapter *adapter;
6370 	struct net_device *netdev;
6371 	unsigned char *mac_addr_p;
6372 	unsigned long flags;
6373 	bool init_success;
6374 	int rc;
6375 
6376 	dev_dbg(&dev->dev, "entering ibmvnic_probe for UA 0x%x\n",
6377 		dev->unit_address);
6378 
6379 	mac_addr_p = (unsigned char *)vio_get_attribute(dev,
6380 							VETH_MAC_ADDR, NULL);
6381 	if (!mac_addr_p) {
6382 		dev_err(&dev->dev,
6383 			"(%s:%3.3d) ERROR: Can't find MAC_ADDR attribute\n",
6384 			__FILE__, __LINE__);
6385 		return 0;
6386 	}
6387 
6388 	netdev = alloc_etherdev_mq(sizeof(struct ibmvnic_adapter),
6389 				   IBMVNIC_MAX_QUEUES);
6390 	if (!netdev)
6391 		return -ENOMEM;
6392 
6393 	adapter = netdev_priv(netdev);
6394 	adapter->state = VNIC_PROBING;
6395 	dev_set_drvdata(&dev->dev, netdev);
6396 	adapter->vdev = dev;
6397 	adapter->netdev = netdev;
6398 	adapter->login_pending = false;
6399 	memset(&adapter->map_ids, 0, sizeof(adapter->map_ids));
6400 	/* map_ids start at 1, so ensure map_id 0 is always "in-use" */
6401 	bitmap_set(adapter->map_ids, 0, 1);
6402 
6403 	ether_addr_copy(adapter->mac_addr, mac_addr_p);
6404 	eth_hw_addr_set(netdev, adapter->mac_addr);
6405 	netdev->irq = dev->irq;
6406 	netdev->netdev_ops = &ibmvnic_netdev_ops;
6407 	netdev->ethtool_ops = &ibmvnic_ethtool_ops;
6408 	SET_NETDEV_DEV(netdev, &dev->dev);
6409 
6410 	INIT_WORK(&adapter->ibmvnic_reset, __ibmvnic_reset);
6411 	INIT_DELAYED_WORK(&adapter->ibmvnic_delayed_reset,
6412 			  __ibmvnic_delayed_reset);
6413 	INIT_LIST_HEAD(&adapter->rwi_list);
6414 	spin_lock_init(&adapter->rwi_lock);
6415 	spin_lock_init(&adapter->state_lock);
6416 	mutex_init(&adapter->fw_lock);
6417 	init_completion(&adapter->probe_done);
6418 	init_completion(&adapter->init_done);
6419 	init_completion(&adapter->fw_done);
6420 	init_completion(&adapter->reset_done);
6421 	init_completion(&adapter->stats_done);
6422 	clear_bit(0, &adapter->resetting);
6423 	adapter->prev_rx_buf_sz = 0;
6424 	adapter->prev_mtu = 0;
6425 
6426 	init_success = false;
6427 	do {
6428 		reinit_init_done(adapter);
6429 
6430 		/* clear any failovers we got in the previous pass
6431 		 * since we are reinitializing the CRQ
6432 		 */
6433 		adapter->failover_pending = false;
6434 
6435 		/* If we had already initialized CRQ, we may have one or
6436 		 * more resets queued already. Discard those and release
6437 		 * the CRQ before initializing the CRQ again.
6438 		 */
6439 		release_crq_queue(adapter);
6440 
6441 		/* Since we are still in PROBING state, __ibmvnic_reset()
6442 		 * will not access the ->rwi_list and since we released CRQ,
6443 		 * we won't get _new_ transport events. But there maybe an
6444 		 * ongoing ibmvnic_reset() call. So serialize access to
6445 		 * rwi_list. If we win the race, ibvmnic_reset() could add
6446 		 * a reset after we purged but thats ok - we just may end
6447 		 * up with an extra reset (i.e similar to having two or more
6448 		 * resets in the queue at once).
6449 		 * CHECK.
6450 		 */
6451 		spin_lock_irqsave(&adapter->rwi_lock, flags);
6452 		flush_reset_queue(adapter);
6453 		spin_unlock_irqrestore(&adapter->rwi_lock, flags);
6454 
6455 		rc = init_crq_queue(adapter);
6456 		if (rc) {
6457 			dev_err(&dev->dev, "Couldn't initialize crq. rc=%d\n",
6458 				rc);
6459 			goto ibmvnic_init_fail;
6460 		}
6461 
6462 		rc = ibmvnic_reset_init(adapter, false);
6463 	} while (rc == -EAGAIN);
6464 
6465 	/* We are ignoring the error from ibmvnic_reset_init() assuming that the
6466 	 * partner is not ready. CRQ is not active. When the partner becomes
6467 	 * ready, we will do the passive init reset.
6468 	 */
6469 
6470 	if (!rc)
6471 		init_success = true;
6472 
6473 	rc = init_stats_buffers(adapter);
6474 	if (rc)
6475 		goto ibmvnic_init_fail;
6476 
6477 	rc = init_stats_token(adapter);
6478 	if (rc)
6479 		goto ibmvnic_stats_fail;
6480 
6481 	rc = device_create_file(&dev->dev, &dev_attr_failover);
6482 	if (rc)
6483 		goto ibmvnic_dev_file_err;
6484 
6485 	netif_carrier_off(netdev);
6486 
6487 	if (init_success) {
6488 		adapter->state = VNIC_PROBED;
6489 		netdev->mtu = adapter->req_mtu - ETH_HLEN;
6490 		netdev->min_mtu = adapter->min_mtu - ETH_HLEN;
6491 		netdev->max_mtu = adapter->max_mtu - ETH_HLEN;
6492 	} else {
6493 		adapter->state = VNIC_DOWN;
6494 	}
6495 
6496 	adapter->wait_for_reset = false;
6497 	adapter->last_reset_time = jiffies;
6498 
6499 	rc = register_netdev(netdev);
6500 	if (rc) {
6501 		dev_err(&dev->dev, "failed to register netdev rc=%d\n", rc);
6502 		goto ibmvnic_register_fail;
6503 	}
6504 	dev_info(&dev->dev, "ibmvnic registered\n");
6505 
6506 	rc = ibmvnic_cpu_notif_add(adapter);
6507 	if (rc) {
6508 		netdev_err(netdev, "Registering cpu notifier failed\n");
6509 		goto cpu_notif_add_failed;
6510 	}
6511 
6512 	complete(&adapter->probe_done);
6513 
6514 	return 0;
6515 
6516 cpu_notif_add_failed:
6517 	unregister_netdev(netdev);
6518 
6519 ibmvnic_register_fail:
6520 	device_remove_file(&dev->dev, &dev_attr_failover);
6521 
6522 ibmvnic_dev_file_err:
6523 	release_stats_token(adapter);
6524 
6525 ibmvnic_stats_fail:
6526 	release_stats_buffers(adapter);
6527 
6528 ibmvnic_init_fail:
6529 	release_sub_crqs(adapter, 1);
6530 	release_crq_queue(adapter);
6531 
6532 	/* cleanup worker thread after releasing CRQ so we don't get
6533 	 * transport events (i.e new work items for the worker thread).
6534 	 */
6535 	adapter->state = VNIC_REMOVING;
6536 	complete(&adapter->probe_done);
6537 	flush_work(&adapter->ibmvnic_reset);
6538 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6539 
6540 	flush_reset_queue(adapter);
6541 
6542 	mutex_destroy(&adapter->fw_lock);
6543 	free_netdev(netdev);
6544 
6545 	return rc;
6546 }
6547 
6548 static void ibmvnic_remove(struct vio_dev *dev)
6549 {
6550 	struct net_device *netdev = dev_get_drvdata(&dev->dev);
6551 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6552 	unsigned long flags;
6553 
6554 	spin_lock_irqsave(&adapter->state_lock, flags);
6555 
6556 	/* If ibmvnic_reset() is scheduling a reset, wait for it to
6557 	 * finish. Then, set the state to REMOVING to prevent it from
6558 	 * scheduling any more work and to have reset functions ignore
6559 	 * any resets that have already been scheduled. Drop the lock
6560 	 * after setting state, so __ibmvnic_reset() which is called
6561 	 * from the flush_work() below, can make progress.
6562 	 */
6563 	spin_lock(&adapter->rwi_lock);
6564 	adapter->state = VNIC_REMOVING;
6565 	spin_unlock(&adapter->rwi_lock);
6566 
6567 	spin_unlock_irqrestore(&adapter->state_lock, flags);
6568 
6569 	ibmvnic_cpu_notif_remove(adapter);
6570 
6571 	flush_work(&adapter->ibmvnic_reset);
6572 	flush_delayed_work(&adapter->ibmvnic_delayed_reset);
6573 
6574 	rtnl_lock();
6575 	unregister_netdevice(netdev);
6576 
6577 	release_resources(adapter);
6578 	release_rx_pools(adapter);
6579 	release_tx_pools(adapter);
6580 	release_sub_crqs(adapter, 1);
6581 	release_crq_queue(adapter);
6582 
6583 	release_stats_token(adapter);
6584 	release_stats_buffers(adapter);
6585 
6586 	adapter->state = VNIC_REMOVED;
6587 
6588 	rtnl_unlock();
6589 	mutex_destroy(&adapter->fw_lock);
6590 	device_remove_file(&dev->dev, &dev_attr_failover);
6591 	free_netdev(netdev);
6592 	dev_set_drvdata(&dev->dev, NULL);
6593 }
6594 
6595 static ssize_t failover_store(struct device *dev, struct device_attribute *attr,
6596 			      const char *buf, size_t count)
6597 {
6598 	struct net_device *netdev = dev_get_drvdata(dev);
6599 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6600 	unsigned long retbuf[PLPAR_HCALL_BUFSIZE];
6601 	__be64 session_token;
6602 	long rc;
6603 
6604 	if (!sysfs_streq(buf, "1"))
6605 		return -EINVAL;
6606 
6607 	rc = plpar_hcall(H_VIOCTL, retbuf, adapter->vdev->unit_address,
6608 			 H_GET_SESSION_TOKEN, 0, 0, 0);
6609 	if (rc) {
6610 		netdev_err(netdev, "Couldn't retrieve session token, rc %ld\n",
6611 			   rc);
6612 		goto last_resort;
6613 	}
6614 
6615 	session_token = (__be64)retbuf[0];
6616 	netdev_dbg(netdev, "Initiating client failover, session id %llx\n",
6617 		   be64_to_cpu(session_token));
6618 	rc = plpar_hcall_norets(H_VIOCTL, adapter->vdev->unit_address,
6619 				H_SESSION_ERR_DETECTED, session_token, 0, 0);
6620 	if (rc) {
6621 		netdev_err(netdev,
6622 			   "H_VIOCTL initiated failover failed, rc %ld\n",
6623 			   rc);
6624 		goto last_resort;
6625 	}
6626 
6627 	return count;
6628 
6629 last_resort:
6630 	netdev_dbg(netdev, "Trying to send CRQ_CMD, the last resort\n");
6631 	ibmvnic_reset(adapter, VNIC_RESET_FAILOVER);
6632 
6633 	return count;
6634 }
6635 static DEVICE_ATTR_WO(failover);
6636 
6637 static unsigned long ibmvnic_get_desired_dma(struct vio_dev *vdev)
6638 {
6639 	struct net_device *netdev = dev_get_drvdata(&vdev->dev);
6640 	struct ibmvnic_adapter *adapter;
6641 	struct iommu_table *tbl;
6642 	unsigned long ret = 0;
6643 	int i;
6644 
6645 	tbl = get_iommu_table_base(&vdev->dev);
6646 
6647 	/* netdev inits at probe time along with the structures we need below*/
6648 	if (!netdev)
6649 		return IOMMU_PAGE_ALIGN(IBMVNIC_IO_ENTITLEMENT_DEFAULT, tbl);
6650 
6651 	adapter = netdev_priv(netdev);
6652 
6653 	ret += PAGE_SIZE; /* the crq message queue */
6654 	ret += IOMMU_PAGE_ALIGN(sizeof(struct ibmvnic_statistics), tbl);
6655 
6656 	for (i = 0; i < adapter->req_tx_queues + adapter->req_rx_queues; i++)
6657 		ret += 4 * PAGE_SIZE; /* the scrq message queue */
6658 
6659 	for (i = 0; i < adapter->num_active_rx_pools; i++)
6660 		ret += adapter->rx_pool[i].size *
6661 		    IOMMU_PAGE_ALIGN(adapter->rx_pool[i].buff_size, tbl);
6662 
6663 	return ret;
6664 }
6665 
6666 static int ibmvnic_resume(struct device *dev)
6667 {
6668 	struct net_device *netdev = dev_get_drvdata(dev);
6669 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
6670 
6671 	if (adapter->state != VNIC_OPEN)
6672 		return 0;
6673 
6674 	tasklet_schedule(&adapter->tasklet);
6675 
6676 	return 0;
6677 }
6678 
6679 static const struct vio_device_id ibmvnic_device_table[] = {
6680 	{"network", "IBM,vnic"},
6681 	{"", "" }
6682 };
6683 MODULE_DEVICE_TABLE(vio, ibmvnic_device_table);
6684 
6685 static const struct dev_pm_ops ibmvnic_pm_ops = {
6686 	.resume = ibmvnic_resume
6687 };
6688 
6689 static struct vio_driver ibmvnic_driver = {
6690 	.id_table       = ibmvnic_device_table,
6691 	.probe          = ibmvnic_probe,
6692 	.remove         = ibmvnic_remove,
6693 	.get_desired_dma = ibmvnic_get_desired_dma,
6694 	.name		= ibmvnic_driver_name,
6695 	.pm		= &ibmvnic_pm_ops,
6696 };
6697 
6698 /* module functions */
6699 static int __init ibmvnic_module_init(void)
6700 {
6701 	int ret;
6702 
6703 	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "net/ibmvnic:online",
6704 				      ibmvnic_cpu_online,
6705 				      ibmvnic_cpu_down_prep);
6706 	if (ret < 0)
6707 		goto out;
6708 	ibmvnic_online = ret;
6709 	ret = cpuhp_setup_state_multi(CPUHP_IBMVNIC_DEAD, "net/ibmvnic:dead",
6710 				      NULL, ibmvnic_cpu_dead);
6711 	if (ret)
6712 		goto err_dead;
6713 
6714 	ret = vio_register_driver(&ibmvnic_driver);
6715 	if (ret)
6716 		goto err_vio_register;
6717 
6718 	pr_info("%s: %s %s\n", ibmvnic_driver_name, ibmvnic_driver_string,
6719 		IBMVNIC_DRIVER_VERSION);
6720 
6721 	return 0;
6722 err_vio_register:
6723 	cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD);
6724 err_dead:
6725 	cpuhp_remove_multi_state(ibmvnic_online);
6726 out:
6727 	return ret;
6728 }
6729 
6730 static void __exit ibmvnic_module_exit(void)
6731 {
6732 	vio_unregister_driver(&ibmvnic_driver);
6733 	cpuhp_remove_multi_state(CPUHP_IBMVNIC_DEAD);
6734 	cpuhp_remove_multi_state(ibmvnic_online);
6735 }
6736 
6737 module_init(ibmvnic_module_init);
6738 module_exit(ibmvnic_module_exit);
6739